xref: /openbmc/linux/arch/powerpc/kvm/book3s_hv.c (revision c8dbaa22)
1 /*
2  * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
3  * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
4  *
5  * Authors:
6  *    Paul Mackerras <paulus@au1.ibm.com>
7  *    Alexander Graf <agraf@suse.de>
8  *    Kevin Wolf <mail@kevin-wolf.de>
9  *
10  * Description: KVM functions specific to running on Book 3S
11  * processors in hypervisor mode (specifically POWER7 and later).
12  *
13  * This file is derived from arch/powerpc/kvm/book3s.c,
14  * by Alexander Graf <agraf@suse.de>.
15  *
16  * This program is free software; you can redistribute it and/or modify
17  * it under the terms of the GNU General Public License, version 2, as
18  * published by the Free Software Foundation.
19  */
20 
21 #include <linux/kvm_host.h>
22 #include <linux/err.h>
23 #include <linux/slab.h>
24 #include <linux/preempt.h>
25 #include <linux/sched/signal.h>
26 #include <linux/sched/stat.h>
27 #include <linux/delay.h>
28 #include <linux/export.h>
29 #include <linux/fs.h>
30 #include <linux/anon_inodes.h>
31 #include <linux/cpu.h>
32 #include <linux/cpumask.h>
33 #include <linux/spinlock.h>
34 #include <linux/page-flags.h>
35 #include <linux/srcu.h>
36 #include <linux/miscdevice.h>
37 #include <linux/debugfs.h>
38 #include <linux/gfp.h>
39 #include <linux/vmalloc.h>
40 #include <linux/highmem.h>
41 #include <linux/hugetlb.h>
42 #include <linux/kvm_irqfd.h>
43 #include <linux/irqbypass.h>
44 #include <linux/module.h>
45 #include <linux/compiler.h>
46 #include <linux/of.h>
47 
48 #include <asm/reg.h>
49 #include <asm/ppc-opcode.h>
50 #include <asm/disassemble.h>
51 #include <asm/cputable.h>
52 #include <asm/cacheflush.h>
53 #include <asm/tlbflush.h>
54 #include <linux/uaccess.h>
55 #include <asm/io.h>
56 #include <asm/kvm_ppc.h>
57 #include <asm/kvm_book3s.h>
58 #include <asm/mmu_context.h>
59 #include <asm/lppaca.h>
60 #include <asm/processor.h>
61 #include <asm/cputhreads.h>
62 #include <asm/page.h>
63 #include <asm/hvcall.h>
64 #include <asm/switch_to.h>
65 #include <asm/smp.h>
66 #include <asm/dbell.h>
67 #include <asm/hmi.h>
68 #include <asm/pnv-pci.h>
69 #include <asm/mmu.h>
70 #include <asm/opal.h>
71 #include <asm/xics.h>
72 #include <asm/xive.h>
73 
74 #include "book3s.h"
75 
76 #define CREATE_TRACE_POINTS
77 #include "trace_hv.h"
78 
79 /* #define EXIT_DEBUG */
80 /* #define EXIT_DEBUG_SIMPLE */
81 /* #define EXIT_DEBUG_INT */
82 
83 /* Used to indicate that a guest page fault needs to be handled */
84 #define RESUME_PAGE_FAULT	(RESUME_GUEST | RESUME_FLAG_ARCH1)
85 /* Used to indicate that a guest passthrough interrupt needs to be handled */
86 #define RESUME_PASSTHROUGH	(RESUME_GUEST | RESUME_FLAG_ARCH2)
87 
88 /* Used as a "null" value for timebase values */
89 #define TB_NIL	(~(u64)0)
90 
91 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
92 
93 static int dynamic_mt_modes = 6;
94 module_param(dynamic_mt_modes, int, S_IRUGO | S_IWUSR);
95 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
96 static int target_smt_mode;
97 module_param(target_smt_mode, int, S_IRUGO | S_IWUSR);
98 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
99 
100 #ifdef CONFIG_KVM_XICS
101 static struct kernel_param_ops module_param_ops = {
102 	.set = param_set_int,
103 	.get = param_get_int,
104 };
105 
106 module_param_cb(kvm_irq_bypass, &module_param_ops, &kvm_irq_bypass,
107 							S_IRUGO | S_IWUSR);
108 MODULE_PARM_DESC(kvm_irq_bypass, "Bypass passthrough interrupt optimization");
109 
110 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect,
111 							S_IRUGO | S_IWUSR);
112 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
113 #endif
114 
115 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
116 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
117 
118 static inline struct kvm_vcpu *next_runnable_thread(struct kvmppc_vcore *vc,
119 		int *ip)
120 {
121 	int i = *ip;
122 	struct kvm_vcpu *vcpu;
123 
124 	while (++i < MAX_SMT_THREADS) {
125 		vcpu = READ_ONCE(vc->runnable_threads[i]);
126 		if (vcpu) {
127 			*ip = i;
128 			return vcpu;
129 		}
130 	}
131 	return NULL;
132 }
133 
134 /* Used to traverse the list of runnable threads for a given vcore */
135 #define for_each_runnable_thread(i, vcpu, vc) \
136 	for (i = -1; (vcpu = next_runnable_thread(vc, &i)); )
137 
138 static bool kvmppc_ipi_thread(int cpu)
139 {
140 	unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
141 
142 	/* On POWER9 we can use msgsnd to IPI any cpu */
143 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
144 		msg |= get_hard_smp_processor_id(cpu);
145 		smp_mb();
146 		__asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
147 		return true;
148 	}
149 
150 	/* On POWER8 for IPIs to threads in the same core, use msgsnd */
151 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
152 		preempt_disable();
153 		if (cpu_first_thread_sibling(cpu) ==
154 		    cpu_first_thread_sibling(smp_processor_id())) {
155 			msg |= cpu_thread_in_core(cpu);
156 			smp_mb();
157 			__asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
158 			preempt_enable();
159 			return true;
160 		}
161 		preempt_enable();
162 	}
163 
164 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
165 	if (cpu >= 0 && cpu < nr_cpu_ids) {
166 		if (paca[cpu].kvm_hstate.xics_phys) {
167 			xics_wake_cpu(cpu);
168 			return true;
169 		}
170 		opal_int_set_mfrr(get_hard_smp_processor_id(cpu), IPI_PRIORITY);
171 		return true;
172 	}
173 #endif
174 
175 	return false;
176 }
177 
178 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
179 {
180 	int cpu;
181 	struct swait_queue_head *wqp;
182 
183 	wqp = kvm_arch_vcpu_wq(vcpu);
184 	if (swait_active(wqp)) {
185 		swake_up(wqp);
186 		++vcpu->stat.halt_wakeup;
187 	}
188 
189 	cpu = READ_ONCE(vcpu->arch.thread_cpu);
190 	if (cpu >= 0 && kvmppc_ipi_thread(cpu))
191 		return;
192 
193 	/* CPU points to the first thread of the core */
194 	cpu = vcpu->cpu;
195 	if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
196 		smp_send_reschedule(cpu);
197 }
198 
199 /*
200  * We use the vcpu_load/put functions to measure stolen time.
201  * Stolen time is counted as time when either the vcpu is able to
202  * run as part of a virtual core, but the task running the vcore
203  * is preempted or sleeping, or when the vcpu needs something done
204  * in the kernel by the task running the vcpu, but that task is
205  * preempted or sleeping.  Those two things have to be counted
206  * separately, since one of the vcpu tasks will take on the job
207  * of running the core, and the other vcpu tasks in the vcore will
208  * sleep waiting for it to do that, but that sleep shouldn't count
209  * as stolen time.
210  *
211  * Hence we accumulate stolen time when the vcpu can run as part of
212  * a vcore using vc->stolen_tb, and the stolen time when the vcpu
213  * needs its task to do other things in the kernel (for example,
214  * service a page fault) in busy_stolen.  We don't accumulate
215  * stolen time for a vcore when it is inactive, or for a vcpu
216  * when it is in state RUNNING or NOTREADY.  NOTREADY is a bit of
217  * a misnomer; it means that the vcpu task is not executing in
218  * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
219  * the kernel.  We don't have any way of dividing up that time
220  * between time that the vcpu is genuinely stopped, time that
221  * the task is actively working on behalf of the vcpu, and time
222  * that the task is preempted, so we don't count any of it as
223  * stolen.
224  *
225  * Updates to busy_stolen are protected by arch.tbacct_lock;
226  * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
227  * lock.  The stolen times are measured in units of timebase ticks.
228  * (Note that the != TB_NIL checks below are purely defensive;
229  * they should never fail.)
230  */
231 
232 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc)
233 {
234 	unsigned long flags;
235 
236 	spin_lock_irqsave(&vc->stoltb_lock, flags);
237 	vc->preempt_tb = mftb();
238 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
239 }
240 
241 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc)
242 {
243 	unsigned long flags;
244 
245 	spin_lock_irqsave(&vc->stoltb_lock, flags);
246 	if (vc->preempt_tb != TB_NIL) {
247 		vc->stolen_tb += mftb() - vc->preempt_tb;
248 		vc->preempt_tb = TB_NIL;
249 	}
250 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
251 }
252 
253 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
254 {
255 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
256 	unsigned long flags;
257 
258 	/*
259 	 * We can test vc->runner without taking the vcore lock,
260 	 * because only this task ever sets vc->runner to this
261 	 * vcpu, and once it is set to this vcpu, only this task
262 	 * ever sets it to NULL.
263 	 */
264 	if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
265 		kvmppc_core_end_stolen(vc);
266 
267 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
268 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
269 	    vcpu->arch.busy_preempt != TB_NIL) {
270 		vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
271 		vcpu->arch.busy_preempt = TB_NIL;
272 	}
273 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
274 }
275 
276 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
277 {
278 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
279 	unsigned long flags;
280 
281 	if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
282 		kvmppc_core_start_stolen(vc);
283 
284 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
285 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
286 		vcpu->arch.busy_preempt = mftb();
287 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
288 }
289 
290 static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
291 {
292 	/*
293 	 * Check for illegal transactional state bit combination
294 	 * and if we find it, force the TS field to a safe state.
295 	 */
296 	if ((msr & MSR_TS_MASK) == MSR_TS_MASK)
297 		msr &= ~MSR_TS_MASK;
298 	vcpu->arch.shregs.msr = msr;
299 	kvmppc_end_cede(vcpu);
300 }
301 
302 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
303 {
304 	vcpu->arch.pvr = pvr;
305 }
306 
307 /* Dummy value used in computing PCR value below */
308 #define PCR_ARCH_300	(PCR_ARCH_207 << 1)
309 
310 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
311 {
312 	unsigned long host_pcr_bit = 0, guest_pcr_bit = 0;
313 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
314 
315 	/* We can (emulate) our own architecture version and anything older */
316 	if (cpu_has_feature(CPU_FTR_ARCH_300))
317 		host_pcr_bit = PCR_ARCH_300;
318 	else if (cpu_has_feature(CPU_FTR_ARCH_207S))
319 		host_pcr_bit = PCR_ARCH_207;
320 	else if (cpu_has_feature(CPU_FTR_ARCH_206))
321 		host_pcr_bit = PCR_ARCH_206;
322 	else
323 		host_pcr_bit = PCR_ARCH_205;
324 
325 	/* Determine lowest PCR bit needed to run guest in given PVR level */
326 	guest_pcr_bit = host_pcr_bit;
327 	if (arch_compat) {
328 		switch (arch_compat) {
329 		case PVR_ARCH_205:
330 			guest_pcr_bit = PCR_ARCH_205;
331 			break;
332 		case PVR_ARCH_206:
333 		case PVR_ARCH_206p:
334 			guest_pcr_bit = PCR_ARCH_206;
335 			break;
336 		case PVR_ARCH_207:
337 			guest_pcr_bit = PCR_ARCH_207;
338 			break;
339 		case PVR_ARCH_300:
340 			guest_pcr_bit = PCR_ARCH_300;
341 			break;
342 		default:
343 			return -EINVAL;
344 		}
345 	}
346 
347 	/* Check requested PCR bits don't exceed our capabilities */
348 	if (guest_pcr_bit > host_pcr_bit)
349 		return -EINVAL;
350 
351 	spin_lock(&vc->lock);
352 	vc->arch_compat = arch_compat;
353 	/* Set all PCR bits for which guest_pcr_bit <= bit < host_pcr_bit */
354 	vc->pcr = host_pcr_bit - guest_pcr_bit;
355 	spin_unlock(&vc->lock);
356 
357 	return 0;
358 }
359 
360 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
361 {
362 	int r;
363 
364 	pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
365 	pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
366 	       vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
367 	for (r = 0; r < 16; ++r)
368 		pr_err("r%2d = %.16lx  r%d = %.16lx\n",
369 		       r, kvmppc_get_gpr(vcpu, r),
370 		       r+16, kvmppc_get_gpr(vcpu, r+16));
371 	pr_err("ctr = %.16lx  lr  = %.16lx\n",
372 	       vcpu->arch.ctr, vcpu->arch.lr);
373 	pr_err("srr0 = %.16llx srr1 = %.16llx\n",
374 	       vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
375 	pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
376 	       vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
377 	pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
378 	       vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
379 	pr_err("cr = %.8x  xer = %.16lx  dsisr = %.8x\n",
380 	       vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
381 	pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
382 	pr_err("fault dar = %.16lx dsisr = %.8x\n",
383 	       vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
384 	pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
385 	for (r = 0; r < vcpu->arch.slb_max; ++r)
386 		pr_err("  ESID = %.16llx VSID = %.16llx\n",
387 		       vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
388 	pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
389 	       vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
390 	       vcpu->arch.last_inst);
391 }
392 
393 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
394 {
395 	struct kvm_vcpu *ret;
396 
397 	mutex_lock(&kvm->lock);
398 	ret = kvm_get_vcpu_by_id(kvm, id);
399 	mutex_unlock(&kvm->lock);
400 	return ret;
401 }
402 
403 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
404 {
405 	vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
406 	vpa->yield_count = cpu_to_be32(1);
407 }
408 
409 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
410 		   unsigned long addr, unsigned long len)
411 {
412 	/* check address is cacheline aligned */
413 	if (addr & (L1_CACHE_BYTES - 1))
414 		return -EINVAL;
415 	spin_lock(&vcpu->arch.vpa_update_lock);
416 	if (v->next_gpa != addr || v->len != len) {
417 		v->next_gpa = addr;
418 		v->len = addr ? len : 0;
419 		v->update_pending = 1;
420 	}
421 	spin_unlock(&vcpu->arch.vpa_update_lock);
422 	return 0;
423 }
424 
425 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
426 struct reg_vpa {
427 	u32 dummy;
428 	union {
429 		__be16 hword;
430 		__be32 word;
431 	} length;
432 };
433 
434 static int vpa_is_registered(struct kvmppc_vpa *vpap)
435 {
436 	if (vpap->update_pending)
437 		return vpap->next_gpa != 0;
438 	return vpap->pinned_addr != NULL;
439 }
440 
441 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
442 				       unsigned long flags,
443 				       unsigned long vcpuid, unsigned long vpa)
444 {
445 	struct kvm *kvm = vcpu->kvm;
446 	unsigned long len, nb;
447 	void *va;
448 	struct kvm_vcpu *tvcpu;
449 	int err;
450 	int subfunc;
451 	struct kvmppc_vpa *vpap;
452 
453 	tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
454 	if (!tvcpu)
455 		return H_PARAMETER;
456 
457 	subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
458 	if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
459 	    subfunc == H_VPA_REG_SLB) {
460 		/* Registering new area - address must be cache-line aligned */
461 		if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
462 			return H_PARAMETER;
463 
464 		/* convert logical addr to kernel addr and read length */
465 		va = kvmppc_pin_guest_page(kvm, vpa, &nb);
466 		if (va == NULL)
467 			return H_PARAMETER;
468 		if (subfunc == H_VPA_REG_VPA)
469 			len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
470 		else
471 			len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
472 		kvmppc_unpin_guest_page(kvm, va, vpa, false);
473 
474 		/* Check length */
475 		if (len > nb || len < sizeof(struct reg_vpa))
476 			return H_PARAMETER;
477 	} else {
478 		vpa = 0;
479 		len = 0;
480 	}
481 
482 	err = H_PARAMETER;
483 	vpap = NULL;
484 	spin_lock(&tvcpu->arch.vpa_update_lock);
485 
486 	switch (subfunc) {
487 	case H_VPA_REG_VPA:		/* register VPA */
488 		if (len < sizeof(struct lppaca))
489 			break;
490 		vpap = &tvcpu->arch.vpa;
491 		err = 0;
492 		break;
493 
494 	case H_VPA_REG_DTL:		/* register DTL */
495 		if (len < sizeof(struct dtl_entry))
496 			break;
497 		len -= len % sizeof(struct dtl_entry);
498 
499 		/* Check that they have previously registered a VPA */
500 		err = H_RESOURCE;
501 		if (!vpa_is_registered(&tvcpu->arch.vpa))
502 			break;
503 
504 		vpap = &tvcpu->arch.dtl;
505 		err = 0;
506 		break;
507 
508 	case H_VPA_REG_SLB:		/* register SLB shadow buffer */
509 		/* Check that they have previously registered a VPA */
510 		err = H_RESOURCE;
511 		if (!vpa_is_registered(&tvcpu->arch.vpa))
512 			break;
513 
514 		vpap = &tvcpu->arch.slb_shadow;
515 		err = 0;
516 		break;
517 
518 	case H_VPA_DEREG_VPA:		/* deregister VPA */
519 		/* Check they don't still have a DTL or SLB buf registered */
520 		err = H_RESOURCE;
521 		if (vpa_is_registered(&tvcpu->arch.dtl) ||
522 		    vpa_is_registered(&tvcpu->arch.slb_shadow))
523 			break;
524 
525 		vpap = &tvcpu->arch.vpa;
526 		err = 0;
527 		break;
528 
529 	case H_VPA_DEREG_DTL:		/* deregister DTL */
530 		vpap = &tvcpu->arch.dtl;
531 		err = 0;
532 		break;
533 
534 	case H_VPA_DEREG_SLB:		/* deregister SLB shadow buffer */
535 		vpap = &tvcpu->arch.slb_shadow;
536 		err = 0;
537 		break;
538 	}
539 
540 	if (vpap) {
541 		vpap->next_gpa = vpa;
542 		vpap->len = len;
543 		vpap->update_pending = 1;
544 	}
545 
546 	spin_unlock(&tvcpu->arch.vpa_update_lock);
547 
548 	return err;
549 }
550 
551 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
552 {
553 	struct kvm *kvm = vcpu->kvm;
554 	void *va;
555 	unsigned long nb;
556 	unsigned long gpa;
557 
558 	/*
559 	 * We need to pin the page pointed to by vpap->next_gpa,
560 	 * but we can't call kvmppc_pin_guest_page under the lock
561 	 * as it does get_user_pages() and down_read().  So we
562 	 * have to drop the lock, pin the page, then get the lock
563 	 * again and check that a new area didn't get registered
564 	 * in the meantime.
565 	 */
566 	for (;;) {
567 		gpa = vpap->next_gpa;
568 		spin_unlock(&vcpu->arch.vpa_update_lock);
569 		va = NULL;
570 		nb = 0;
571 		if (gpa)
572 			va = kvmppc_pin_guest_page(kvm, gpa, &nb);
573 		spin_lock(&vcpu->arch.vpa_update_lock);
574 		if (gpa == vpap->next_gpa)
575 			break;
576 		/* sigh... unpin that one and try again */
577 		if (va)
578 			kvmppc_unpin_guest_page(kvm, va, gpa, false);
579 	}
580 
581 	vpap->update_pending = 0;
582 	if (va && nb < vpap->len) {
583 		/*
584 		 * If it's now too short, it must be that userspace
585 		 * has changed the mappings underlying guest memory,
586 		 * so unregister the region.
587 		 */
588 		kvmppc_unpin_guest_page(kvm, va, gpa, false);
589 		va = NULL;
590 	}
591 	if (vpap->pinned_addr)
592 		kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
593 					vpap->dirty);
594 	vpap->gpa = gpa;
595 	vpap->pinned_addr = va;
596 	vpap->dirty = false;
597 	if (va)
598 		vpap->pinned_end = va + vpap->len;
599 }
600 
601 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
602 {
603 	if (!(vcpu->arch.vpa.update_pending ||
604 	      vcpu->arch.slb_shadow.update_pending ||
605 	      vcpu->arch.dtl.update_pending))
606 		return;
607 
608 	spin_lock(&vcpu->arch.vpa_update_lock);
609 	if (vcpu->arch.vpa.update_pending) {
610 		kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
611 		if (vcpu->arch.vpa.pinned_addr)
612 			init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
613 	}
614 	if (vcpu->arch.dtl.update_pending) {
615 		kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
616 		vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
617 		vcpu->arch.dtl_index = 0;
618 	}
619 	if (vcpu->arch.slb_shadow.update_pending)
620 		kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
621 	spin_unlock(&vcpu->arch.vpa_update_lock);
622 }
623 
624 /*
625  * Return the accumulated stolen time for the vcore up until `now'.
626  * The caller should hold the vcore lock.
627  */
628 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
629 {
630 	u64 p;
631 	unsigned long flags;
632 
633 	spin_lock_irqsave(&vc->stoltb_lock, flags);
634 	p = vc->stolen_tb;
635 	if (vc->vcore_state != VCORE_INACTIVE &&
636 	    vc->preempt_tb != TB_NIL)
637 		p += now - vc->preempt_tb;
638 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
639 	return p;
640 }
641 
642 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
643 				    struct kvmppc_vcore *vc)
644 {
645 	struct dtl_entry *dt;
646 	struct lppaca *vpa;
647 	unsigned long stolen;
648 	unsigned long core_stolen;
649 	u64 now;
650 	unsigned long flags;
651 
652 	dt = vcpu->arch.dtl_ptr;
653 	vpa = vcpu->arch.vpa.pinned_addr;
654 	now = mftb();
655 	core_stolen = vcore_stolen_time(vc, now);
656 	stolen = core_stolen - vcpu->arch.stolen_logged;
657 	vcpu->arch.stolen_logged = core_stolen;
658 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
659 	stolen += vcpu->arch.busy_stolen;
660 	vcpu->arch.busy_stolen = 0;
661 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
662 	if (!dt || !vpa)
663 		return;
664 	memset(dt, 0, sizeof(struct dtl_entry));
665 	dt->dispatch_reason = 7;
666 	dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
667 	dt->timebase = cpu_to_be64(now + vc->tb_offset);
668 	dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
669 	dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
670 	dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
671 	++dt;
672 	if (dt == vcpu->arch.dtl.pinned_end)
673 		dt = vcpu->arch.dtl.pinned_addr;
674 	vcpu->arch.dtl_ptr = dt;
675 	/* order writing *dt vs. writing vpa->dtl_idx */
676 	smp_wmb();
677 	vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
678 	vcpu->arch.dtl.dirty = true;
679 }
680 
681 /* See if there is a doorbell interrupt pending for a vcpu */
682 static bool kvmppc_doorbell_pending(struct kvm_vcpu *vcpu)
683 {
684 	int thr;
685 	struct kvmppc_vcore *vc;
686 
687 	if (vcpu->arch.doorbell_request)
688 		return true;
689 	/*
690 	 * Ensure that the read of vcore->dpdes comes after the read
691 	 * of vcpu->doorbell_request.  This barrier matches the
692 	 * lwsync in book3s_hv_rmhandlers.S just before the
693 	 * fast_guest_return label.
694 	 */
695 	smp_rmb();
696 	vc = vcpu->arch.vcore;
697 	thr = vcpu->vcpu_id - vc->first_vcpuid;
698 	return !!(vc->dpdes & (1 << thr));
699 }
700 
701 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
702 {
703 	if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
704 		return true;
705 	if ((!vcpu->arch.vcore->arch_compat) &&
706 	    cpu_has_feature(CPU_FTR_ARCH_207S))
707 		return true;
708 	return false;
709 }
710 
711 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
712 			     unsigned long resource, unsigned long value1,
713 			     unsigned long value2)
714 {
715 	switch (resource) {
716 	case H_SET_MODE_RESOURCE_SET_CIABR:
717 		if (!kvmppc_power8_compatible(vcpu))
718 			return H_P2;
719 		if (value2)
720 			return H_P4;
721 		if (mflags)
722 			return H_UNSUPPORTED_FLAG_START;
723 		/* Guests can't breakpoint the hypervisor */
724 		if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
725 			return H_P3;
726 		vcpu->arch.ciabr  = value1;
727 		return H_SUCCESS;
728 	case H_SET_MODE_RESOURCE_SET_DAWR:
729 		if (!kvmppc_power8_compatible(vcpu))
730 			return H_P2;
731 		if (mflags)
732 			return H_UNSUPPORTED_FLAG_START;
733 		if (value2 & DABRX_HYP)
734 			return H_P4;
735 		vcpu->arch.dawr  = value1;
736 		vcpu->arch.dawrx = value2;
737 		return H_SUCCESS;
738 	default:
739 		return H_TOO_HARD;
740 	}
741 }
742 
743 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
744 {
745 	struct kvmppc_vcore *vcore = target->arch.vcore;
746 
747 	/*
748 	 * We expect to have been called by the real mode handler
749 	 * (kvmppc_rm_h_confer()) which would have directly returned
750 	 * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
751 	 * have useful work to do and should not confer) so we don't
752 	 * recheck that here.
753 	 */
754 
755 	spin_lock(&vcore->lock);
756 	if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
757 	    vcore->vcore_state != VCORE_INACTIVE &&
758 	    vcore->runner)
759 		target = vcore->runner;
760 	spin_unlock(&vcore->lock);
761 
762 	return kvm_vcpu_yield_to(target);
763 }
764 
765 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
766 {
767 	int yield_count = 0;
768 	struct lppaca *lppaca;
769 
770 	spin_lock(&vcpu->arch.vpa_update_lock);
771 	lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
772 	if (lppaca)
773 		yield_count = be32_to_cpu(lppaca->yield_count);
774 	spin_unlock(&vcpu->arch.vpa_update_lock);
775 	return yield_count;
776 }
777 
778 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
779 {
780 	unsigned long req = kvmppc_get_gpr(vcpu, 3);
781 	unsigned long target, ret = H_SUCCESS;
782 	int yield_count;
783 	struct kvm_vcpu *tvcpu;
784 	int idx, rc;
785 
786 	if (req <= MAX_HCALL_OPCODE &&
787 	    !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
788 		return RESUME_HOST;
789 
790 	switch (req) {
791 	case H_CEDE:
792 		break;
793 	case H_PROD:
794 		target = kvmppc_get_gpr(vcpu, 4);
795 		tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
796 		if (!tvcpu) {
797 			ret = H_PARAMETER;
798 			break;
799 		}
800 		tvcpu->arch.prodded = 1;
801 		smp_mb();
802 		if (tvcpu->arch.ceded)
803 			kvmppc_fast_vcpu_kick_hv(tvcpu);
804 		break;
805 	case H_CONFER:
806 		target = kvmppc_get_gpr(vcpu, 4);
807 		if (target == -1)
808 			break;
809 		tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
810 		if (!tvcpu) {
811 			ret = H_PARAMETER;
812 			break;
813 		}
814 		yield_count = kvmppc_get_gpr(vcpu, 5);
815 		if (kvmppc_get_yield_count(tvcpu) != yield_count)
816 			break;
817 		kvm_arch_vcpu_yield_to(tvcpu);
818 		break;
819 	case H_REGISTER_VPA:
820 		ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
821 					kvmppc_get_gpr(vcpu, 5),
822 					kvmppc_get_gpr(vcpu, 6));
823 		break;
824 	case H_RTAS:
825 		if (list_empty(&vcpu->kvm->arch.rtas_tokens))
826 			return RESUME_HOST;
827 
828 		idx = srcu_read_lock(&vcpu->kvm->srcu);
829 		rc = kvmppc_rtas_hcall(vcpu);
830 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
831 
832 		if (rc == -ENOENT)
833 			return RESUME_HOST;
834 		else if (rc == 0)
835 			break;
836 
837 		/* Send the error out to userspace via KVM_RUN */
838 		return rc;
839 	case H_LOGICAL_CI_LOAD:
840 		ret = kvmppc_h_logical_ci_load(vcpu);
841 		if (ret == H_TOO_HARD)
842 			return RESUME_HOST;
843 		break;
844 	case H_LOGICAL_CI_STORE:
845 		ret = kvmppc_h_logical_ci_store(vcpu);
846 		if (ret == H_TOO_HARD)
847 			return RESUME_HOST;
848 		break;
849 	case H_SET_MODE:
850 		ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
851 					kvmppc_get_gpr(vcpu, 5),
852 					kvmppc_get_gpr(vcpu, 6),
853 					kvmppc_get_gpr(vcpu, 7));
854 		if (ret == H_TOO_HARD)
855 			return RESUME_HOST;
856 		break;
857 	case H_XIRR:
858 	case H_CPPR:
859 	case H_EOI:
860 	case H_IPI:
861 	case H_IPOLL:
862 	case H_XIRR_X:
863 		if (kvmppc_xics_enabled(vcpu)) {
864 			if (xive_enabled()) {
865 				ret = H_NOT_AVAILABLE;
866 				return RESUME_GUEST;
867 			}
868 			ret = kvmppc_xics_hcall(vcpu, req);
869 			break;
870 		}
871 		return RESUME_HOST;
872 	case H_PUT_TCE:
873 		ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
874 						kvmppc_get_gpr(vcpu, 5),
875 						kvmppc_get_gpr(vcpu, 6));
876 		if (ret == H_TOO_HARD)
877 			return RESUME_HOST;
878 		break;
879 	case H_PUT_TCE_INDIRECT:
880 		ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
881 						kvmppc_get_gpr(vcpu, 5),
882 						kvmppc_get_gpr(vcpu, 6),
883 						kvmppc_get_gpr(vcpu, 7));
884 		if (ret == H_TOO_HARD)
885 			return RESUME_HOST;
886 		break;
887 	case H_STUFF_TCE:
888 		ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
889 						kvmppc_get_gpr(vcpu, 5),
890 						kvmppc_get_gpr(vcpu, 6),
891 						kvmppc_get_gpr(vcpu, 7));
892 		if (ret == H_TOO_HARD)
893 			return RESUME_HOST;
894 		break;
895 	default:
896 		return RESUME_HOST;
897 	}
898 	kvmppc_set_gpr(vcpu, 3, ret);
899 	vcpu->arch.hcall_needed = 0;
900 	return RESUME_GUEST;
901 }
902 
903 static int kvmppc_hcall_impl_hv(unsigned long cmd)
904 {
905 	switch (cmd) {
906 	case H_CEDE:
907 	case H_PROD:
908 	case H_CONFER:
909 	case H_REGISTER_VPA:
910 	case H_SET_MODE:
911 	case H_LOGICAL_CI_LOAD:
912 	case H_LOGICAL_CI_STORE:
913 #ifdef CONFIG_KVM_XICS
914 	case H_XIRR:
915 	case H_CPPR:
916 	case H_EOI:
917 	case H_IPI:
918 	case H_IPOLL:
919 	case H_XIRR_X:
920 #endif
921 		return 1;
922 	}
923 
924 	/* See if it's in the real-mode table */
925 	return kvmppc_hcall_impl_hv_realmode(cmd);
926 }
927 
928 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
929 					struct kvm_vcpu *vcpu)
930 {
931 	u32 last_inst;
932 
933 	if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
934 					EMULATE_DONE) {
935 		/*
936 		 * Fetch failed, so return to guest and
937 		 * try executing it again.
938 		 */
939 		return RESUME_GUEST;
940 	}
941 
942 	if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
943 		run->exit_reason = KVM_EXIT_DEBUG;
944 		run->debug.arch.address = kvmppc_get_pc(vcpu);
945 		return RESUME_HOST;
946 	} else {
947 		kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
948 		return RESUME_GUEST;
949 	}
950 }
951 
952 static void do_nothing(void *x)
953 {
954 }
955 
956 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
957 {
958 	int thr, cpu, pcpu, nthreads;
959 	struct kvm_vcpu *v;
960 	unsigned long dpdes;
961 
962 	nthreads = vcpu->kvm->arch.emul_smt_mode;
963 	dpdes = 0;
964 	cpu = vcpu->vcpu_id & ~(nthreads - 1);
965 	for (thr = 0; thr < nthreads; ++thr, ++cpu) {
966 		v = kvmppc_find_vcpu(vcpu->kvm, cpu);
967 		if (!v)
968 			continue;
969 		/*
970 		 * If the vcpu is currently running on a physical cpu thread,
971 		 * interrupt it in order to pull it out of the guest briefly,
972 		 * which will update its vcore->dpdes value.
973 		 */
974 		pcpu = READ_ONCE(v->cpu);
975 		if (pcpu >= 0)
976 			smp_call_function_single(pcpu, do_nothing, NULL, 1);
977 		if (kvmppc_doorbell_pending(v))
978 			dpdes |= 1 << thr;
979 	}
980 	return dpdes;
981 }
982 
983 /*
984  * On POWER9, emulate doorbell-related instructions in order to
985  * give the guest the illusion of running on a multi-threaded core.
986  * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
987  * and mfspr DPDES.
988  */
989 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
990 {
991 	u32 inst, rb, thr;
992 	unsigned long arg;
993 	struct kvm *kvm = vcpu->kvm;
994 	struct kvm_vcpu *tvcpu;
995 
996 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
997 		return EMULATE_FAIL;
998 	if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &inst) != EMULATE_DONE)
999 		return RESUME_GUEST;
1000 	if (get_op(inst) != 31)
1001 		return EMULATE_FAIL;
1002 	rb = get_rb(inst);
1003 	thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
1004 	switch (get_xop(inst)) {
1005 	case OP_31_XOP_MSGSNDP:
1006 		arg = kvmppc_get_gpr(vcpu, rb);
1007 		if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1008 			break;
1009 		arg &= 0x3f;
1010 		if (arg >= kvm->arch.emul_smt_mode)
1011 			break;
1012 		tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
1013 		if (!tvcpu)
1014 			break;
1015 		if (!tvcpu->arch.doorbell_request) {
1016 			tvcpu->arch.doorbell_request = 1;
1017 			kvmppc_fast_vcpu_kick_hv(tvcpu);
1018 		}
1019 		break;
1020 	case OP_31_XOP_MSGCLRP:
1021 		arg = kvmppc_get_gpr(vcpu, rb);
1022 		if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1023 			break;
1024 		vcpu->arch.vcore->dpdes = 0;
1025 		vcpu->arch.doorbell_request = 0;
1026 		break;
1027 	case OP_31_XOP_MFSPR:
1028 		switch (get_sprn(inst)) {
1029 		case SPRN_TIR:
1030 			arg = thr;
1031 			break;
1032 		case SPRN_DPDES:
1033 			arg = kvmppc_read_dpdes(vcpu);
1034 			break;
1035 		default:
1036 			return EMULATE_FAIL;
1037 		}
1038 		kvmppc_set_gpr(vcpu, get_rt(inst), arg);
1039 		break;
1040 	default:
1041 		return EMULATE_FAIL;
1042 	}
1043 	kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
1044 	return RESUME_GUEST;
1045 }
1046 
1047 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
1048 				 struct task_struct *tsk)
1049 {
1050 	int r = RESUME_HOST;
1051 
1052 	vcpu->stat.sum_exits++;
1053 
1054 	/*
1055 	 * This can happen if an interrupt occurs in the last stages
1056 	 * of guest entry or the first stages of guest exit (i.e. after
1057 	 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1058 	 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1059 	 * That can happen due to a bug, or due to a machine check
1060 	 * occurring at just the wrong time.
1061 	 */
1062 	if (vcpu->arch.shregs.msr & MSR_HV) {
1063 		printk(KERN_EMERG "KVM trap in HV mode!\n");
1064 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1065 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
1066 			vcpu->arch.shregs.msr);
1067 		kvmppc_dump_regs(vcpu);
1068 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1069 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1070 		return RESUME_HOST;
1071 	}
1072 	run->exit_reason = KVM_EXIT_UNKNOWN;
1073 	run->ready_for_interrupt_injection = 1;
1074 	switch (vcpu->arch.trap) {
1075 	/* We're good on these - the host merely wanted to get our attention */
1076 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
1077 		vcpu->stat.dec_exits++;
1078 		r = RESUME_GUEST;
1079 		break;
1080 	case BOOK3S_INTERRUPT_EXTERNAL:
1081 	case BOOK3S_INTERRUPT_H_DOORBELL:
1082 	case BOOK3S_INTERRUPT_H_VIRT:
1083 		vcpu->stat.ext_intr_exits++;
1084 		r = RESUME_GUEST;
1085 		break;
1086 	/* HMI is hypervisor interrupt and host has handled it. Resume guest.*/
1087 	case BOOK3S_INTERRUPT_HMI:
1088 	case BOOK3S_INTERRUPT_PERFMON:
1089 		r = RESUME_GUEST;
1090 		break;
1091 	case BOOK3S_INTERRUPT_MACHINE_CHECK:
1092 		/* Exit to guest with KVM_EXIT_NMI as exit reason */
1093 		run->exit_reason = KVM_EXIT_NMI;
1094 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1095 		/* Clear out the old NMI status from run->flags */
1096 		run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK;
1097 		/* Now set the NMI status */
1098 		if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED)
1099 			run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
1100 		else
1101 			run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV;
1102 
1103 		r = RESUME_HOST;
1104 		/* Print the MCE event to host console. */
1105 		machine_check_print_event_info(&vcpu->arch.mce_evt, false);
1106 		break;
1107 	case BOOK3S_INTERRUPT_PROGRAM:
1108 	{
1109 		ulong flags;
1110 		/*
1111 		 * Normally program interrupts are delivered directly
1112 		 * to the guest by the hardware, but we can get here
1113 		 * as a result of a hypervisor emulation interrupt
1114 		 * (e40) getting turned into a 700 by BML RTAS.
1115 		 */
1116 		flags = vcpu->arch.shregs.msr & 0x1f0000ull;
1117 		kvmppc_core_queue_program(vcpu, flags);
1118 		r = RESUME_GUEST;
1119 		break;
1120 	}
1121 	case BOOK3S_INTERRUPT_SYSCALL:
1122 	{
1123 		/* hcall - punt to userspace */
1124 		int i;
1125 
1126 		/* hypercall with MSR_PR has already been handled in rmode,
1127 		 * and never reaches here.
1128 		 */
1129 
1130 		run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1131 		for (i = 0; i < 9; ++i)
1132 			run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1133 		run->exit_reason = KVM_EXIT_PAPR_HCALL;
1134 		vcpu->arch.hcall_needed = 1;
1135 		r = RESUME_HOST;
1136 		break;
1137 	}
1138 	/*
1139 	 * We get these next two if the guest accesses a page which it thinks
1140 	 * it has mapped but which is not actually present, either because
1141 	 * it is for an emulated I/O device or because the corresonding
1142 	 * host page has been paged out.  Any other HDSI/HISI interrupts
1143 	 * have been handled already.
1144 	 */
1145 	case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1146 		r = RESUME_PAGE_FAULT;
1147 		break;
1148 	case BOOK3S_INTERRUPT_H_INST_STORAGE:
1149 		vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1150 		vcpu->arch.fault_dsisr = 0;
1151 		r = RESUME_PAGE_FAULT;
1152 		break;
1153 	/*
1154 	 * This occurs if the guest executes an illegal instruction.
1155 	 * If the guest debug is disabled, generate a program interrupt
1156 	 * to the guest. If guest debug is enabled, we need to check
1157 	 * whether the instruction is a software breakpoint instruction.
1158 	 * Accordingly return to Guest or Host.
1159 	 */
1160 	case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1161 		if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1162 			vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1163 				swab32(vcpu->arch.emul_inst) :
1164 				vcpu->arch.emul_inst;
1165 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1166 			r = kvmppc_emulate_debug_inst(run, vcpu);
1167 		} else {
1168 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1169 			r = RESUME_GUEST;
1170 		}
1171 		break;
1172 	/*
1173 	 * This occurs if the guest (kernel or userspace), does something that
1174 	 * is prohibited by HFSCR.
1175 	 * On POWER9, this could be a doorbell instruction that we need
1176 	 * to emulate.
1177 	 * Otherwise, we just generate a program interrupt to the guest.
1178 	 */
1179 	case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
1180 		r = EMULATE_FAIL;
1181 		if ((vcpu->arch.hfscr >> 56) == FSCR_MSGP_LG)
1182 			r = kvmppc_emulate_doorbell_instr(vcpu);
1183 		if (r == EMULATE_FAIL) {
1184 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1185 			r = RESUME_GUEST;
1186 		}
1187 		break;
1188 	case BOOK3S_INTERRUPT_HV_RM_HARD:
1189 		r = RESUME_PASSTHROUGH;
1190 		break;
1191 	default:
1192 		kvmppc_dump_regs(vcpu);
1193 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1194 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
1195 			vcpu->arch.shregs.msr);
1196 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1197 		r = RESUME_HOST;
1198 		break;
1199 	}
1200 
1201 	return r;
1202 }
1203 
1204 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1205 					    struct kvm_sregs *sregs)
1206 {
1207 	int i;
1208 
1209 	memset(sregs, 0, sizeof(struct kvm_sregs));
1210 	sregs->pvr = vcpu->arch.pvr;
1211 	for (i = 0; i < vcpu->arch.slb_max; i++) {
1212 		sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1213 		sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1214 	}
1215 
1216 	return 0;
1217 }
1218 
1219 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1220 					    struct kvm_sregs *sregs)
1221 {
1222 	int i, j;
1223 
1224 	/* Only accept the same PVR as the host's, since we can't spoof it */
1225 	if (sregs->pvr != vcpu->arch.pvr)
1226 		return -EINVAL;
1227 
1228 	j = 0;
1229 	for (i = 0; i < vcpu->arch.slb_nr; i++) {
1230 		if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1231 			vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1232 			vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1233 			++j;
1234 		}
1235 	}
1236 	vcpu->arch.slb_max = j;
1237 
1238 	return 0;
1239 }
1240 
1241 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1242 		bool preserve_top32)
1243 {
1244 	struct kvm *kvm = vcpu->kvm;
1245 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
1246 	u64 mask;
1247 
1248 	mutex_lock(&kvm->lock);
1249 	spin_lock(&vc->lock);
1250 	/*
1251 	 * If ILE (interrupt little-endian) has changed, update the
1252 	 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1253 	 */
1254 	if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1255 		struct kvm_vcpu *vcpu;
1256 		int i;
1257 
1258 		kvm_for_each_vcpu(i, vcpu, kvm) {
1259 			if (vcpu->arch.vcore != vc)
1260 				continue;
1261 			if (new_lpcr & LPCR_ILE)
1262 				vcpu->arch.intr_msr |= MSR_LE;
1263 			else
1264 				vcpu->arch.intr_msr &= ~MSR_LE;
1265 		}
1266 	}
1267 
1268 	/*
1269 	 * Userspace can only modify DPFD (default prefetch depth),
1270 	 * ILE (interrupt little-endian) and TC (translation control).
1271 	 * On POWER8 and POWER9 userspace can also modify AIL (alt. interrupt loc.).
1272 	 */
1273 	mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1274 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
1275 		mask |= LPCR_AIL;
1276 	/*
1277 	 * On POWER9, allow userspace to enable large decrementer for the
1278 	 * guest, whether or not the host has it enabled.
1279 	 */
1280 	if (cpu_has_feature(CPU_FTR_ARCH_300))
1281 		mask |= LPCR_LD;
1282 
1283 	/* Broken 32-bit version of LPCR must not clear top bits */
1284 	if (preserve_top32)
1285 		mask &= 0xFFFFFFFF;
1286 	vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1287 	spin_unlock(&vc->lock);
1288 	mutex_unlock(&kvm->lock);
1289 }
1290 
1291 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1292 				 union kvmppc_one_reg *val)
1293 {
1294 	int r = 0;
1295 	long int i;
1296 
1297 	switch (id) {
1298 	case KVM_REG_PPC_DEBUG_INST:
1299 		*val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1300 		break;
1301 	case KVM_REG_PPC_HIOR:
1302 		*val = get_reg_val(id, 0);
1303 		break;
1304 	case KVM_REG_PPC_DABR:
1305 		*val = get_reg_val(id, vcpu->arch.dabr);
1306 		break;
1307 	case KVM_REG_PPC_DABRX:
1308 		*val = get_reg_val(id, vcpu->arch.dabrx);
1309 		break;
1310 	case KVM_REG_PPC_DSCR:
1311 		*val = get_reg_val(id, vcpu->arch.dscr);
1312 		break;
1313 	case KVM_REG_PPC_PURR:
1314 		*val = get_reg_val(id, vcpu->arch.purr);
1315 		break;
1316 	case KVM_REG_PPC_SPURR:
1317 		*val = get_reg_val(id, vcpu->arch.spurr);
1318 		break;
1319 	case KVM_REG_PPC_AMR:
1320 		*val = get_reg_val(id, vcpu->arch.amr);
1321 		break;
1322 	case KVM_REG_PPC_UAMOR:
1323 		*val = get_reg_val(id, vcpu->arch.uamor);
1324 		break;
1325 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1326 		i = id - KVM_REG_PPC_MMCR0;
1327 		*val = get_reg_val(id, vcpu->arch.mmcr[i]);
1328 		break;
1329 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1330 		i = id - KVM_REG_PPC_PMC1;
1331 		*val = get_reg_val(id, vcpu->arch.pmc[i]);
1332 		break;
1333 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1334 		i = id - KVM_REG_PPC_SPMC1;
1335 		*val = get_reg_val(id, vcpu->arch.spmc[i]);
1336 		break;
1337 	case KVM_REG_PPC_SIAR:
1338 		*val = get_reg_val(id, vcpu->arch.siar);
1339 		break;
1340 	case KVM_REG_PPC_SDAR:
1341 		*val = get_reg_val(id, vcpu->arch.sdar);
1342 		break;
1343 	case KVM_REG_PPC_SIER:
1344 		*val = get_reg_val(id, vcpu->arch.sier);
1345 		break;
1346 	case KVM_REG_PPC_IAMR:
1347 		*val = get_reg_val(id, vcpu->arch.iamr);
1348 		break;
1349 	case KVM_REG_PPC_PSPB:
1350 		*val = get_reg_val(id, vcpu->arch.pspb);
1351 		break;
1352 	case KVM_REG_PPC_DPDES:
1353 		*val = get_reg_val(id, vcpu->arch.vcore->dpdes);
1354 		break;
1355 	case KVM_REG_PPC_VTB:
1356 		*val = get_reg_val(id, vcpu->arch.vcore->vtb);
1357 		break;
1358 	case KVM_REG_PPC_DAWR:
1359 		*val = get_reg_val(id, vcpu->arch.dawr);
1360 		break;
1361 	case KVM_REG_PPC_DAWRX:
1362 		*val = get_reg_val(id, vcpu->arch.dawrx);
1363 		break;
1364 	case KVM_REG_PPC_CIABR:
1365 		*val = get_reg_val(id, vcpu->arch.ciabr);
1366 		break;
1367 	case KVM_REG_PPC_CSIGR:
1368 		*val = get_reg_val(id, vcpu->arch.csigr);
1369 		break;
1370 	case KVM_REG_PPC_TACR:
1371 		*val = get_reg_val(id, vcpu->arch.tacr);
1372 		break;
1373 	case KVM_REG_PPC_TCSCR:
1374 		*val = get_reg_val(id, vcpu->arch.tcscr);
1375 		break;
1376 	case KVM_REG_PPC_PID:
1377 		*val = get_reg_val(id, vcpu->arch.pid);
1378 		break;
1379 	case KVM_REG_PPC_ACOP:
1380 		*val = get_reg_val(id, vcpu->arch.acop);
1381 		break;
1382 	case KVM_REG_PPC_WORT:
1383 		*val = get_reg_val(id, vcpu->arch.wort);
1384 		break;
1385 	case KVM_REG_PPC_TIDR:
1386 		*val = get_reg_val(id, vcpu->arch.tid);
1387 		break;
1388 	case KVM_REG_PPC_PSSCR:
1389 		*val = get_reg_val(id, vcpu->arch.psscr);
1390 		break;
1391 	case KVM_REG_PPC_VPA_ADDR:
1392 		spin_lock(&vcpu->arch.vpa_update_lock);
1393 		*val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1394 		spin_unlock(&vcpu->arch.vpa_update_lock);
1395 		break;
1396 	case KVM_REG_PPC_VPA_SLB:
1397 		spin_lock(&vcpu->arch.vpa_update_lock);
1398 		val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1399 		val->vpaval.length = vcpu->arch.slb_shadow.len;
1400 		spin_unlock(&vcpu->arch.vpa_update_lock);
1401 		break;
1402 	case KVM_REG_PPC_VPA_DTL:
1403 		spin_lock(&vcpu->arch.vpa_update_lock);
1404 		val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1405 		val->vpaval.length = vcpu->arch.dtl.len;
1406 		spin_unlock(&vcpu->arch.vpa_update_lock);
1407 		break;
1408 	case KVM_REG_PPC_TB_OFFSET:
1409 		*val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1410 		break;
1411 	case KVM_REG_PPC_LPCR:
1412 	case KVM_REG_PPC_LPCR_64:
1413 		*val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1414 		break;
1415 	case KVM_REG_PPC_PPR:
1416 		*val = get_reg_val(id, vcpu->arch.ppr);
1417 		break;
1418 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1419 	case KVM_REG_PPC_TFHAR:
1420 		*val = get_reg_val(id, vcpu->arch.tfhar);
1421 		break;
1422 	case KVM_REG_PPC_TFIAR:
1423 		*val = get_reg_val(id, vcpu->arch.tfiar);
1424 		break;
1425 	case KVM_REG_PPC_TEXASR:
1426 		*val = get_reg_val(id, vcpu->arch.texasr);
1427 		break;
1428 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1429 		i = id - KVM_REG_PPC_TM_GPR0;
1430 		*val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1431 		break;
1432 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1433 	{
1434 		int j;
1435 		i = id - KVM_REG_PPC_TM_VSR0;
1436 		if (i < 32)
1437 			for (j = 0; j < TS_FPRWIDTH; j++)
1438 				val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1439 		else {
1440 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
1441 				val->vval = vcpu->arch.vr_tm.vr[i-32];
1442 			else
1443 				r = -ENXIO;
1444 		}
1445 		break;
1446 	}
1447 	case KVM_REG_PPC_TM_CR:
1448 		*val = get_reg_val(id, vcpu->arch.cr_tm);
1449 		break;
1450 	case KVM_REG_PPC_TM_XER:
1451 		*val = get_reg_val(id, vcpu->arch.xer_tm);
1452 		break;
1453 	case KVM_REG_PPC_TM_LR:
1454 		*val = get_reg_val(id, vcpu->arch.lr_tm);
1455 		break;
1456 	case KVM_REG_PPC_TM_CTR:
1457 		*val = get_reg_val(id, vcpu->arch.ctr_tm);
1458 		break;
1459 	case KVM_REG_PPC_TM_FPSCR:
1460 		*val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1461 		break;
1462 	case KVM_REG_PPC_TM_AMR:
1463 		*val = get_reg_val(id, vcpu->arch.amr_tm);
1464 		break;
1465 	case KVM_REG_PPC_TM_PPR:
1466 		*val = get_reg_val(id, vcpu->arch.ppr_tm);
1467 		break;
1468 	case KVM_REG_PPC_TM_VRSAVE:
1469 		*val = get_reg_val(id, vcpu->arch.vrsave_tm);
1470 		break;
1471 	case KVM_REG_PPC_TM_VSCR:
1472 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
1473 			*val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1474 		else
1475 			r = -ENXIO;
1476 		break;
1477 	case KVM_REG_PPC_TM_DSCR:
1478 		*val = get_reg_val(id, vcpu->arch.dscr_tm);
1479 		break;
1480 	case KVM_REG_PPC_TM_TAR:
1481 		*val = get_reg_val(id, vcpu->arch.tar_tm);
1482 		break;
1483 #endif
1484 	case KVM_REG_PPC_ARCH_COMPAT:
1485 		*val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1486 		break;
1487 	default:
1488 		r = -EINVAL;
1489 		break;
1490 	}
1491 
1492 	return r;
1493 }
1494 
1495 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1496 				 union kvmppc_one_reg *val)
1497 {
1498 	int r = 0;
1499 	long int i;
1500 	unsigned long addr, len;
1501 
1502 	switch (id) {
1503 	case KVM_REG_PPC_HIOR:
1504 		/* Only allow this to be set to zero */
1505 		if (set_reg_val(id, *val))
1506 			r = -EINVAL;
1507 		break;
1508 	case KVM_REG_PPC_DABR:
1509 		vcpu->arch.dabr = set_reg_val(id, *val);
1510 		break;
1511 	case KVM_REG_PPC_DABRX:
1512 		vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1513 		break;
1514 	case KVM_REG_PPC_DSCR:
1515 		vcpu->arch.dscr = set_reg_val(id, *val);
1516 		break;
1517 	case KVM_REG_PPC_PURR:
1518 		vcpu->arch.purr = set_reg_val(id, *val);
1519 		break;
1520 	case KVM_REG_PPC_SPURR:
1521 		vcpu->arch.spurr = set_reg_val(id, *val);
1522 		break;
1523 	case KVM_REG_PPC_AMR:
1524 		vcpu->arch.amr = set_reg_val(id, *val);
1525 		break;
1526 	case KVM_REG_PPC_UAMOR:
1527 		vcpu->arch.uamor = set_reg_val(id, *val);
1528 		break;
1529 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1530 		i = id - KVM_REG_PPC_MMCR0;
1531 		vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1532 		break;
1533 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1534 		i = id - KVM_REG_PPC_PMC1;
1535 		vcpu->arch.pmc[i] = set_reg_val(id, *val);
1536 		break;
1537 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1538 		i = id - KVM_REG_PPC_SPMC1;
1539 		vcpu->arch.spmc[i] = set_reg_val(id, *val);
1540 		break;
1541 	case KVM_REG_PPC_SIAR:
1542 		vcpu->arch.siar = set_reg_val(id, *val);
1543 		break;
1544 	case KVM_REG_PPC_SDAR:
1545 		vcpu->arch.sdar = set_reg_val(id, *val);
1546 		break;
1547 	case KVM_REG_PPC_SIER:
1548 		vcpu->arch.sier = set_reg_val(id, *val);
1549 		break;
1550 	case KVM_REG_PPC_IAMR:
1551 		vcpu->arch.iamr = set_reg_val(id, *val);
1552 		break;
1553 	case KVM_REG_PPC_PSPB:
1554 		vcpu->arch.pspb = set_reg_val(id, *val);
1555 		break;
1556 	case KVM_REG_PPC_DPDES:
1557 		vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1558 		break;
1559 	case KVM_REG_PPC_VTB:
1560 		vcpu->arch.vcore->vtb = set_reg_val(id, *val);
1561 		break;
1562 	case KVM_REG_PPC_DAWR:
1563 		vcpu->arch.dawr = set_reg_val(id, *val);
1564 		break;
1565 	case KVM_REG_PPC_DAWRX:
1566 		vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1567 		break;
1568 	case KVM_REG_PPC_CIABR:
1569 		vcpu->arch.ciabr = set_reg_val(id, *val);
1570 		/* Don't allow setting breakpoints in hypervisor code */
1571 		if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1572 			vcpu->arch.ciabr &= ~CIABR_PRIV;	/* disable */
1573 		break;
1574 	case KVM_REG_PPC_CSIGR:
1575 		vcpu->arch.csigr = set_reg_val(id, *val);
1576 		break;
1577 	case KVM_REG_PPC_TACR:
1578 		vcpu->arch.tacr = set_reg_val(id, *val);
1579 		break;
1580 	case KVM_REG_PPC_TCSCR:
1581 		vcpu->arch.tcscr = set_reg_val(id, *val);
1582 		break;
1583 	case KVM_REG_PPC_PID:
1584 		vcpu->arch.pid = set_reg_val(id, *val);
1585 		break;
1586 	case KVM_REG_PPC_ACOP:
1587 		vcpu->arch.acop = set_reg_val(id, *val);
1588 		break;
1589 	case KVM_REG_PPC_WORT:
1590 		vcpu->arch.wort = set_reg_val(id, *val);
1591 		break;
1592 	case KVM_REG_PPC_TIDR:
1593 		vcpu->arch.tid = set_reg_val(id, *val);
1594 		break;
1595 	case KVM_REG_PPC_PSSCR:
1596 		vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
1597 		break;
1598 	case KVM_REG_PPC_VPA_ADDR:
1599 		addr = set_reg_val(id, *val);
1600 		r = -EINVAL;
1601 		if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1602 			      vcpu->arch.dtl.next_gpa))
1603 			break;
1604 		r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1605 		break;
1606 	case KVM_REG_PPC_VPA_SLB:
1607 		addr = val->vpaval.addr;
1608 		len = val->vpaval.length;
1609 		r = -EINVAL;
1610 		if (addr && !vcpu->arch.vpa.next_gpa)
1611 			break;
1612 		r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1613 		break;
1614 	case KVM_REG_PPC_VPA_DTL:
1615 		addr = val->vpaval.addr;
1616 		len = val->vpaval.length;
1617 		r = -EINVAL;
1618 		if (addr && (len < sizeof(struct dtl_entry) ||
1619 			     !vcpu->arch.vpa.next_gpa))
1620 			break;
1621 		len -= len % sizeof(struct dtl_entry);
1622 		r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1623 		break;
1624 	case KVM_REG_PPC_TB_OFFSET:
1625 		/*
1626 		 * POWER9 DD1 has an erratum where writing TBU40 causes
1627 		 * the timebase to lose ticks.  So we don't let the
1628 		 * timebase offset be changed on P9 DD1.  (It is
1629 		 * initialized to zero.)
1630 		 */
1631 		if (cpu_has_feature(CPU_FTR_POWER9_DD1))
1632 			break;
1633 		/* round up to multiple of 2^24 */
1634 		vcpu->arch.vcore->tb_offset =
1635 			ALIGN(set_reg_val(id, *val), 1UL << 24);
1636 		break;
1637 	case KVM_REG_PPC_LPCR:
1638 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1639 		break;
1640 	case KVM_REG_PPC_LPCR_64:
1641 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1642 		break;
1643 	case KVM_REG_PPC_PPR:
1644 		vcpu->arch.ppr = set_reg_val(id, *val);
1645 		break;
1646 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1647 	case KVM_REG_PPC_TFHAR:
1648 		vcpu->arch.tfhar = set_reg_val(id, *val);
1649 		break;
1650 	case KVM_REG_PPC_TFIAR:
1651 		vcpu->arch.tfiar = set_reg_val(id, *val);
1652 		break;
1653 	case KVM_REG_PPC_TEXASR:
1654 		vcpu->arch.texasr = set_reg_val(id, *val);
1655 		break;
1656 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1657 		i = id - KVM_REG_PPC_TM_GPR0;
1658 		vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1659 		break;
1660 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1661 	{
1662 		int j;
1663 		i = id - KVM_REG_PPC_TM_VSR0;
1664 		if (i < 32)
1665 			for (j = 0; j < TS_FPRWIDTH; j++)
1666 				vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1667 		else
1668 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
1669 				vcpu->arch.vr_tm.vr[i-32] = val->vval;
1670 			else
1671 				r = -ENXIO;
1672 		break;
1673 	}
1674 	case KVM_REG_PPC_TM_CR:
1675 		vcpu->arch.cr_tm = set_reg_val(id, *val);
1676 		break;
1677 	case KVM_REG_PPC_TM_XER:
1678 		vcpu->arch.xer_tm = set_reg_val(id, *val);
1679 		break;
1680 	case KVM_REG_PPC_TM_LR:
1681 		vcpu->arch.lr_tm = set_reg_val(id, *val);
1682 		break;
1683 	case KVM_REG_PPC_TM_CTR:
1684 		vcpu->arch.ctr_tm = set_reg_val(id, *val);
1685 		break;
1686 	case KVM_REG_PPC_TM_FPSCR:
1687 		vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1688 		break;
1689 	case KVM_REG_PPC_TM_AMR:
1690 		vcpu->arch.amr_tm = set_reg_val(id, *val);
1691 		break;
1692 	case KVM_REG_PPC_TM_PPR:
1693 		vcpu->arch.ppr_tm = set_reg_val(id, *val);
1694 		break;
1695 	case KVM_REG_PPC_TM_VRSAVE:
1696 		vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1697 		break;
1698 	case KVM_REG_PPC_TM_VSCR:
1699 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
1700 			vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1701 		else
1702 			r = - ENXIO;
1703 		break;
1704 	case KVM_REG_PPC_TM_DSCR:
1705 		vcpu->arch.dscr_tm = set_reg_val(id, *val);
1706 		break;
1707 	case KVM_REG_PPC_TM_TAR:
1708 		vcpu->arch.tar_tm = set_reg_val(id, *val);
1709 		break;
1710 #endif
1711 	case KVM_REG_PPC_ARCH_COMPAT:
1712 		r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1713 		break;
1714 	default:
1715 		r = -EINVAL;
1716 		break;
1717 	}
1718 
1719 	return r;
1720 }
1721 
1722 /*
1723  * On POWER9, threads are independent and can be in different partitions.
1724  * Therefore we consider each thread to be a subcore.
1725  * There is a restriction that all threads have to be in the same
1726  * MMU mode (radix or HPT), unfortunately, but since we only support
1727  * HPT guests on a HPT host so far, that isn't an impediment yet.
1728  */
1729 static int threads_per_vcore(void)
1730 {
1731 	if (cpu_has_feature(CPU_FTR_ARCH_300))
1732 		return 1;
1733 	return threads_per_subcore;
1734 }
1735 
1736 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int core)
1737 {
1738 	struct kvmppc_vcore *vcore;
1739 
1740 	vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1741 
1742 	if (vcore == NULL)
1743 		return NULL;
1744 
1745 	spin_lock_init(&vcore->lock);
1746 	spin_lock_init(&vcore->stoltb_lock);
1747 	init_swait_queue_head(&vcore->wq);
1748 	vcore->preempt_tb = TB_NIL;
1749 	vcore->lpcr = kvm->arch.lpcr;
1750 	vcore->first_vcpuid = core * kvm->arch.smt_mode;
1751 	vcore->kvm = kvm;
1752 	INIT_LIST_HEAD(&vcore->preempt_list);
1753 
1754 	return vcore;
1755 }
1756 
1757 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1758 static struct debugfs_timings_element {
1759 	const char *name;
1760 	size_t offset;
1761 } timings[] = {
1762 	{"rm_entry",	offsetof(struct kvm_vcpu, arch.rm_entry)},
1763 	{"rm_intr",	offsetof(struct kvm_vcpu, arch.rm_intr)},
1764 	{"rm_exit",	offsetof(struct kvm_vcpu, arch.rm_exit)},
1765 	{"guest",	offsetof(struct kvm_vcpu, arch.guest_time)},
1766 	{"cede",	offsetof(struct kvm_vcpu, arch.cede_time)},
1767 };
1768 
1769 #define N_TIMINGS	(sizeof(timings) / sizeof(timings[0]))
1770 
1771 struct debugfs_timings_state {
1772 	struct kvm_vcpu	*vcpu;
1773 	unsigned int	buflen;
1774 	char		buf[N_TIMINGS * 100];
1775 };
1776 
1777 static int debugfs_timings_open(struct inode *inode, struct file *file)
1778 {
1779 	struct kvm_vcpu *vcpu = inode->i_private;
1780 	struct debugfs_timings_state *p;
1781 
1782 	p = kzalloc(sizeof(*p), GFP_KERNEL);
1783 	if (!p)
1784 		return -ENOMEM;
1785 
1786 	kvm_get_kvm(vcpu->kvm);
1787 	p->vcpu = vcpu;
1788 	file->private_data = p;
1789 
1790 	return nonseekable_open(inode, file);
1791 }
1792 
1793 static int debugfs_timings_release(struct inode *inode, struct file *file)
1794 {
1795 	struct debugfs_timings_state *p = file->private_data;
1796 
1797 	kvm_put_kvm(p->vcpu->kvm);
1798 	kfree(p);
1799 	return 0;
1800 }
1801 
1802 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
1803 				    size_t len, loff_t *ppos)
1804 {
1805 	struct debugfs_timings_state *p = file->private_data;
1806 	struct kvm_vcpu *vcpu = p->vcpu;
1807 	char *s, *buf_end;
1808 	struct kvmhv_tb_accumulator tb;
1809 	u64 count;
1810 	loff_t pos;
1811 	ssize_t n;
1812 	int i, loops;
1813 	bool ok;
1814 
1815 	if (!p->buflen) {
1816 		s = p->buf;
1817 		buf_end = s + sizeof(p->buf);
1818 		for (i = 0; i < N_TIMINGS; ++i) {
1819 			struct kvmhv_tb_accumulator *acc;
1820 
1821 			acc = (struct kvmhv_tb_accumulator *)
1822 				((unsigned long)vcpu + timings[i].offset);
1823 			ok = false;
1824 			for (loops = 0; loops < 1000; ++loops) {
1825 				count = acc->seqcount;
1826 				if (!(count & 1)) {
1827 					smp_rmb();
1828 					tb = *acc;
1829 					smp_rmb();
1830 					if (count == acc->seqcount) {
1831 						ok = true;
1832 						break;
1833 					}
1834 				}
1835 				udelay(1);
1836 			}
1837 			if (!ok)
1838 				snprintf(s, buf_end - s, "%s: stuck\n",
1839 					timings[i].name);
1840 			else
1841 				snprintf(s, buf_end - s,
1842 					"%s: %llu %llu %llu %llu\n",
1843 					timings[i].name, count / 2,
1844 					tb_to_ns(tb.tb_total),
1845 					tb_to_ns(tb.tb_min),
1846 					tb_to_ns(tb.tb_max));
1847 			s += strlen(s);
1848 		}
1849 		p->buflen = s - p->buf;
1850 	}
1851 
1852 	pos = *ppos;
1853 	if (pos >= p->buflen)
1854 		return 0;
1855 	if (len > p->buflen - pos)
1856 		len = p->buflen - pos;
1857 	n = copy_to_user(buf, p->buf + pos, len);
1858 	if (n) {
1859 		if (n == len)
1860 			return -EFAULT;
1861 		len -= n;
1862 	}
1863 	*ppos = pos + len;
1864 	return len;
1865 }
1866 
1867 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
1868 				     size_t len, loff_t *ppos)
1869 {
1870 	return -EACCES;
1871 }
1872 
1873 static const struct file_operations debugfs_timings_ops = {
1874 	.owner	 = THIS_MODULE,
1875 	.open	 = debugfs_timings_open,
1876 	.release = debugfs_timings_release,
1877 	.read	 = debugfs_timings_read,
1878 	.write	 = debugfs_timings_write,
1879 	.llseek	 = generic_file_llseek,
1880 };
1881 
1882 /* Create a debugfs directory for the vcpu */
1883 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1884 {
1885 	char buf[16];
1886 	struct kvm *kvm = vcpu->kvm;
1887 
1888 	snprintf(buf, sizeof(buf), "vcpu%u", id);
1889 	if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
1890 		return;
1891 	vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
1892 	if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
1893 		return;
1894 	vcpu->arch.debugfs_timings =
1895 		debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
1896 				    vcpu, &debugfs_timings_ops);
1897 }
1898 
1899 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1900 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1901 {
1902 }
1903 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1904 
1905 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1906 						   unsigned int id)
1907 {
1908 	struct kvm_vcpu *vcpu;
1909 	int err;
1910 	int core;
1911 	struct kvmppc_vcore *vcore;
1912 
1913 	err = -ENOMEM;
1914 	vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1915 	if (!vcpu)
1916 		goto out;
1917 
1918 	err = kvm_vcpu_init(vcpu, kvm, id);
1919 	if (err)
1920 		goto free_vcpu;
1921 
1922 	vcpu->arch.shared = &vcpu->arch.shregs;
1923 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1924 	/*
1925 	 * The shared struct is never shared on HV,
1926 	 * so we can always use host endianness
1927 	 */
1928 #ifdef __BIG_ENDIAN__
1929 	vcpu->arch.shared_big_endian = true;
1930 #else
1931 	vcpu->arch.shared_big_endian = false;
1932 #endif
1933 #endif
1934 	vcpu->arch.mmcr[0] = MMCR0_FC;
1935 	vcpu->arch.ctrl = CTRL_RUNLATCH;
1936 	/* default to host PVR, since we can't spoof it */
1937 	kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
1938 	spin_lock_init(&vcpu->arch.vpa_update_lock);
1939 	spin_lock_init(&vcpu->arch.tbacct_lock);
1940 	vcpu->arch.busy_preempt = TB_NIL;
1941 	vcpu->arch.intr_msr = MSR_SF | MSR_ME;
1942 
1943 	/*
1944 	 * Set the default HFSCR for the guest from the host value.
1945 	 * This value is only used on POWER9.
1946 	 * On POWER9 DD1, TM doesn't work, so we make sure to
1947 	 * prevent the guest from using it.
1948 	 * On POWER9, we want to virtualize the doorbell facility, so we
1949 	 * turn off the HFSCR bit, which causes those instructions to trap.
1950 	 */
1951 	vcpu->arch.hfscr = mfspr(SPRN_HFSCR);
1952 	if (!cpu_has_feature(CPU_FTR_TM))
1953 		vcpu->arch.hfscr &= ~HFSCR_TM;
1954 	if (cpu_has_feature(CPU_FTR_ARCH_300))
1955 		vcpu->arch.hfscr &= ~HFSCR_MSGP;
1956 
1957 	kvmppc_mmu_book3s_hv_init(vcpu);
1958 
1959 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1960 
1961 	init_waitqueue_head(&vcpu->arch.cpu_run);
1962 
1963 	mutex_lock(&kvm->lock);
1964 	vcore = NULL;
1965 	err = -EINVAL;
1966 	core = id / kvm->arch.smt_mode;
1967 	if (core < KVM_MAX_VCORES) {
1968 		vcore = kvm->arch.vcores[core];
1969 		if (!vcore) {
1970 			err = -ENOMEM;
1971 			vcore = kvmppc_vcore_create(kvm, core);
1972 			kvm->arch.vcores[core] = vcore;
1973 			kvm->arch.online_vcores++;
1974 		}
1975 	}
1976 	mutex_unlock(&kvm->lock);
1977 
1978 	if (!vcore)
1979 		goto free_vcpu;
1980 
1981 	spin_lock(&vcore->lock);
1982 	++vcore->num_threads;
1983 	spin_unlock(&vcore->lock);
1984 	vcpu->arch.vcore = vcore;
1985 	vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
1986 	vcpu->arch.thread_cpu = -1;
1987 	vcpu->arch.prev_cpu = -1;
1988 
1989 	vcpu->arch.cpu_type = KVM_CPU_3S_64;
1990 	kvmppc_sanity_check(vcpu);
1991 
1992 	debugfs_vcpu_init(vcpu, id);
1993 
1994 	return vcpu;
1995 
1996 free_vcpu:
1997 	kmem_cache_free(kvm_vcpu_cache, vcpu);
1998 out:
1999 	return ERR_PTR(err);
2000 }
2001 
2002 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
2003 			      unsigned long flags)
2004 {
2005 	int err;
2006 	int esmt = 0;
2007 
2008 	if (flags)
2009 		return -EINVAL;
2010 	if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
2011 		return -EINVAL;
2012 	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
2013 		/*
2014 		 * On POWER8 (or POWER7), the threading mode is "strict",
2015 		 * so we pack smt_mode vcpus per vcore.
2016 		 */
2017 		if (smt_mode > threads_per_subcore)
2018 			return -EINVAL;
2019 	} else {
2020 		/*
2021 		 * On POWER9, the threading mode is "loose",
2022 		 * so each vcpu gets its own vcore.
2023 		 */
2024 		esmt = smt_mode;
2025 		smt_mode = 1;
2026 	}
2027 	mutex_lock(&kvm->lock);
2028 	err = -EBUSY;
2029 	if (!kvm->arch.online_vcores) {
2030 		kvm->arch.smt_mode = smt_mode;
2031 		kvm->arch.emul_smt_mode = esmt;
2032 		err = 0;
2033 	}
2034 	mutex_unlock(&kvm->lock);
2035 
2036 	return err;
2037 }
2038 
2039 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
2040 {
2041 	if (vpa->pinned_addr)
2042 		kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
2043 					vpa->dirty);
2044 }
2045 
2046 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
2047 {
2048 	spin_lock(&vcpu->arch.vpa_update_lock);
2049 	unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
2050 	unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
2051 	unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2052 	spin_unlock(&vcpu->arch.vpa_update_lock);
2053 	kvm_vcpu_uninit(vcpu);
2054 	kmem_cache_free(kvm_vcpu_cache, vcpu);
2055 }
2056 
2057 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
2058 {
2059 	/* Indicate we want to get back into the guest */
2060 	return 1;
2061 }
2062 
2063 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
2064 {
2065 	unsigned long dec_nsec, now;
2066 
2067 	now = get_tb();
2068 	if (now > vcpu->arch.dec_expires) {
2069 		/* decrementer has already gone negative */
2070 		kvmppc_core_queue_dec(vcpu);
2071 		kvmppc_core_prepare_to_enter(vcpu);
2072 		return;
2073 	}
2074 	dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
2075 		   / tb_ticks_per_sec;
2076 	hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
2077 	vcpu->arch.timer_running = 1;
2078 }
2079 
2080 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
2081 {
2082 	vcpu->arch.ceded = 0;
2083 	if (vcpu->arch.timer_running) {
2084 		hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2085 		vcpu->arch.timer_running = 0;
2086 	}
2087 }
2088 
2089 extern int __kvmppc_vcore_entry(void);
2090 
2091 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
2092 				   struct kvm_vcpu *vcpu)
2093 {
2094 	u64 now;
2095 
2096 	if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2097 		return;
2098 	spin_lock_irq(&vcpu->arch.tbacct_lock);
2099 	now = mftb();
2100 	vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
2101 		vcpu->arch.stolen_logged;
2102 	vcpu->arch.busy_preempt = now;
2103 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2104 	spin_unlock_irq(&vcpu->arch.tbacct_lock);
2105 	--vc->n_runnable;
2106 	WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
2107 }
2108 
2109 static int kvmppc_grab_hwthread(int cpu)
2110 {
2111 	struct paca_struct *tpaca;
2112 	long timeout = 10000;
2113 
2114 	tpaca = &paca[cpu];
2115 
2116 	/* Ensure the thread won't go into the kernel if it wakes */
2117 	tpaca->kvm_hstate.kvm_vcpu = NULL;
2118 	tpaca->kvm_hstate.kvm_vcore = NULL;
2119 	tpaca->kvm_hstate.napping = 0;
2120 	smp_wmb();
2121 	tpaca->kvm_hstate.hwthread_req = 1;
2122 
2123 	/*
2124 	 * If the thread is already executing in the kernel (e.g. handling
2125 	 * a stray interrupt), wait for it to get back to nap mode.
2126 	 * The smp_mb() is to ensure that our setting of hwthread_req
2127 	 * is visible before we look at hwthread_state, so if this
2128 	 * races with the code at system_reset_pSeries and the thread
2129 	 * misses our setting of hwthread_req, we are sure to see its
2130 	 * setting of hwthread_state, and vice versa.
2131 	 */
2132 	smp_mb();
2133 	while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
2134 		if (--timeout <= 0) {
2135 			pr_err("KVM: couldn't grab cpu %d\n", cpu);
2136 			return -EBUSY;
2137 		}
2138 		udelay(1);
2139 	}
2140 	return 0;
2141 }
2142 
2143 static void kvmppc_release_hwthread(int cpu)
2144 {
2145 	struct paca_struct *tpaca;
2146 
2147 	tpaca = &paca[cpu];
2148 	tpaca->kvm_hstate.hwthread_req = 0;
2149 	tpaca->kvm_hstate.kvm_vcpu = NULL;
2150 	tpaca->kvm_hstate.kvm_vcore = NULL;
2151 	tpaca->kvm_hstate.kvm_split_mode = NULL;
2152 }
2153 
2154 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
2155 {
2156 	int i;
2157 
2158 	cpu = cpu_first_thread_sibling(cpu);
2159 	cpumask_set_cpu(cpu, &kvm->arch.need_tlb_flush);
2160 	/*
2161 	 * Make sure setting of bit in need_tlb_flush precedes
2162 	 * testing of cpu_in_guest bits.  The matching barrier on
2163 	 * the other side is the first smp_mb() in kvmppc_run_core().
2164 	 */
2165 	smp_mb();
2166 	for (i = 0; i < threads_per_core; ++i)
2167 		if (cpumask_test_cpu(cpu + i, &kvm->arch.cpu_in_guest))
2168 			smp_call_function_single(cpu + i, do_nothing, NULL, 1);
2169 }
2170 
2171 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
2172 {
2173 	struct kvm *kvm = vcpu->kvm;
2174 
2175 	/*
2176 	 * With radix, the guest can do TLB invalidations itself,
2177 	 * and it could choose to use the local form (tlbiel) if
2178 	 * it is invalidating a translation that has only ever been
2179 	 * used on one vcpu.  However, that doesn't mean it has
2180 	 * only ever been used on one physical cpu, since vcpus
2181 	 * can move around between pcpus.  To cope with this, when
2182 	 * a vcpu moves from one pcpu to another, we need to tell
2183 	 * any vcpus running on the same core as this vcpu previously
2184 	 * ran to flush the TLB.  The TLB is shared between threads,
2185 	 * so we use a single bit in .need_tlb_flush for all 4 threads.
2186 	 */
2187 	if (vcpu->arch.prev_cpu != pcpu) {
2188 		if (vcpu->arch.prev_cpu >= 0 &&
2189 		    cpu_first_thread_sibling(vcpu->arch.prev_cpu) !=
2190 		    cpu_first_thread_sibling(pcpu))
2191 			radix_flush_cpu(kvm, vcpu->arch.prev_cpu, vcpu);
2192 		vcpu->arch.prev_cpu = pcpu;
2193 	}
2194 }
2195 
2196 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
2197 {
2198 	int cpu;
2199 	struct paca_struct *tpaca;
2200 	struct kvm *kvm = vc->kvm;
2201 
2202 	cpu = vc->pcpu;
2203 	if (vcpu) {
2204 		if (vcpu->arch.timer_running) {
2205 			hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2206 			vcpu->arch.timer_running = 0;
2207 		}
2208 		cpu += vcpu->arch.ptid;
2209 		vcpu->cpu = vc->pcpu;
2210 		vcpu->arch.thread_cpu = cpu;
2211 		cpumask_set_cpu(cpu, &kvm->arch.cpu_in_guest);
2212 	}
2213 	tpaca = &paca[cpu];
2214 	tpaca->kvm_hstate.kvm_vcpu = vcpu;
2215 	tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
2216 	/* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
2217 	smp_wmb();
2218 	tpaca->kvm_hstate.kvm_vcore = vc;
2219 	if (cpu != smp_processor_id())
2220 		kvmppc_ipi_thread(cpu);
2221 }
2222 
2223 static void kvmppc_wait_for_nap(void)
2224 {
2225 	int cpu = smp_processor_id();
2226 	int i, loops;
2227 	int n_threads = threads_per_vcore();
2228 
2229 	if (n_threads <= 1)
2230 		return;
2231 	for (loops = 0; loops < 1000000; ++loops) {
2232 		/*
2233 		 * Check if all threads are finished.
2234 		 * We set the vcore pointer when starting a thread
2235 		 * and the thread clears it when finished, so we look
2236 		 * for any threads that still have a non-NULL vcore ptr.
2237 		 */
2238 		for (i = 1; i < n_threads; ++i)
2239 			if (paca[cpu + i].kvm_hstate.kvm_vcore)
2240 				break;
2241 		if (i == n_threads) {
2242 			HMT_medium();
2243 			return;
2244 		}
2245 		HMT_low();
2246 	}
2247 	HMT_medium();
2248 	for (i = 1; i < n_threads; ++i)
2249 		if (paca[cpu + i].kvm_hstate.kvm_vcore)
2250 			pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
2251 }
2252 
2253 /*
2254  * Check that we are on thread 0 and that any other threads in
2255  * this core are off-line.  Then grab the threads so they can't
2256  * enter the kernel.
2257  */
2258 static int on_primary_thread(void)
2259 {
2260 	int cpu = smp_processor_id();
2261 	int thr;
2262 
2263 	/* Are we on a primary subcore? */
2264 	if (cpu_thread_in_subcore(cpu))
2265 		return 0;
2266 
2267 	thr = 0;
2268 	while (++thr < threads_per_subcore)
2269 		if (cpu_online(cpu + thr))
2270 			return 0;
2271 
2272 	/* Grab all hw threads so they can't go into the kernel */
2273 	for (thr = 1; thr < threads_per_subcore; ++thr) {
2274 		if (kvmppc_grab_hwthread(cpu + thr)) {
2275 			/* Couldn't grab one; let the others go */
2276 			do {
2277 				kvmppc_release_hwthread(cpu + thr);
2278 			} while (--thr > 0);
2279 			return 0;
2280 		}
2281 	}
2282 	return 1;
2283 }
2284 
2285 /*
2286  * A list of virtual cores for each physical CPU.
2287  * These are vcores that could run but their runner VCPU tasks are
2288  * (or may be) preempted.
2289  */
2290 struct preempted_vcore_list {
2291 	struct list_head	list;
2292 	spinlock_t		lock;
2293 };
2294 
2295 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
2296 
2297 static void init_vcore_lists(void)
2298 {
2299 	int cpu;
2300 
2301 	for_each_possible_cpu(cpu) {
2302 		struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
2303 		spin_lock_init(&lp->lock);
2304 		INIT_LIST_HEAD(&lp->list);
2305 	}
2306 }
2307 
2308 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
2309 {
2310 	struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2311 
2312 	vc->vcore_state = VCORE_PREEMPT;
2313 	vc->pcpu = smp_processor_id();
2314 	if (vc->num_threads < threads_per_vcore()) {
2315 		spin_lock(&lp->lock);
2316 		list_add_tail(&vc->preempt_list, &lp->list);
2317 		spin_unlock(&lp->lock);
2318 	}
2319 
2320 	/* Start accumulating stolen time */
2321 	kvmppc_core_start_stolen(vc);
2322 }
2323 
2324 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
2325 {
2326 	struct preempted_vcore_list *lp;
2327 
2328 	kvmppc_core_end_stolen(vc);
2329 	if (!list_empty(&vc->preempt_list)) {
2330 		lp = &per_cpu(preempted_vcores, vc->pcpu);
2331 		spin_lock(&lp->lock);
2332 		list_del_init(&vc->preempt_list);
2333 		spin_unlock(&lp->lock);
2334 	}
2335 	vc->vcore_state = VCORE_INACTIVE;
2336 }
2337 
2338 /*
2339  * This stores information about the virtual cores currently
2340  * assigned to a physical core.
2341  */
2342 struct core_info {
2343 	int		n_subcores;
2344 	int		max_subcore_threads;
2345 	int		total_threads;
2346 	int		subcore_threads[MAX_SUBCORES];
2347 	struct kvmppc_vcore *vc[MAX_SUBCORES];
2348 };
2349 
2350 /*
2351  * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2352  * respectively in 2-way micro-threading (split-core) mode.
2353  */
2354 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2355 
2356 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2357 {
2358 	memset(cip, 0, sizeof(*cip));
2359 	cip->n_subcores = 1;
2360 	cip->max_subcore_threads = vc->num_threads;
2361 	cip->total_threads = vc->num_threads;
2362 	cip->subcore_threads[0] = vc->num_threads;
2363 	cip->vc[0] = vc;
2364 }
2365 
2366 static bool subcore_config_ok(int n_subcores, int n_threads)
2367 {
2368 	/* Can only dynamically split if unsplit to begin with */
2369 	if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2370 		return false;
2371 	if (n_subcores > MAX_SUBCORES)
2372 		return false;
2373 	if (n_subcores > 1) {
2374 		if (!(dynamic_mt_modes & 2))
2375 			n_subcores = 4;
2376 		if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2377 			return false;
2378 	}
2379 
2380 	return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2381 }
2382 
2383 static void init_vcore_to_run(struct kvmppc_vcore *vc)
2384 {
2385 	vc->entry_exit_map = 0;
2386 	vc->in_guest = 0;
2387 	vc->napping_threads = 0;
2388 	vc->conferring_threads = 0;
2389 }
2390 
2391 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2392 {
2393 	int n_threads = vc->num_threads;
2394 	int sub;
2395 
2396 	if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2397 		return false;
2398 
2399 	if (n_threads < cip->max_subcore_threads)
2400 		n_threads = cip->max_subcore_threads;
2401 	if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
2402 		return false;
2403 	cip->max_subcore_threads = n_threads;
2404 
2405 	sub = cip->n_subcores;
2406 	++cip->n_subcores;
2407 	cip->total_threads += vc->num_threads;
2408 	cip->subcore_threads[sub] = vc->num_threads;
2409 	cip->vc[sub] = vc;
2410 	init_vcore_to_run(vc);
2411 	list_del_init(&vc->preempt_list);
2412 
2413 	return true;
2414 }
2415 
2416 /*
2417  * Work out whether it is possible to piggyback the execution of
2418  * vcore *pvc onto the execution of the other vcores described in *cip.
2419  */
2420 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2421 			  int target_threads)
2422 {
2423 	if (cip->total_threads + pvc->num_threads > target_threads)
2424 		return false;
2425 
2426 	return can_dynamic_split(pvc, cip);
2427 }
2428 
2429 static void prepare_threads(struct kvmppc_vcore *vc)
2430 {
2431 	int i;
2432 	struct kvm_vcpu *vcpu;
2433 
2434 	for_each_runnable_thread(i, vcpu, vc) {
2435 		if (signal_pending(vcpu->arch.run_task))
2436 			vcpu->arch.ret = -EINTR;
2437 		else if (vcpu->arch.vpa.update_pending ||
2438 			 vcpu->arch.slb_shadow.update_pending ||
2439 			 vcpu->arch.dtl.update_pending)
2440 			vcpu->arch.ret = RESUME_GUEST;
2441 		else
2442 			continue;
2443 		kvmppc_remove_runnable(vc, vcpu);
2444 		wake_up(&vcpu->arch.cpu_run);
2445 	}
2446 }
2447 
2448 static void collect_piggybacks(struct core_info *cip, int target_threads)
2449 {
2450 	struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2451 	struct kvmppc_vcore *pvc, *vcnext;
2452 
2453 	spin_lock(&lp->lock);
2454 	list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2455 		if (!spin_trylock(&pvc->lock))
2456 			continue;
2457 		prepare_threads(pvc);
2458 		if (!pvc->n_runnable) {
2459 			list_del_init(&pvc->preempt_list);
2460 			if (pvc->runner == NULL) {
2461 				pvc->vcore_state = VCORE_INACTIVE;
2462 				kvmppc_core_end_stolen(pvc);
2463 			}
2464 			spin_unlock(&pvc->lock);
2465 			continue;
2466 		}
2467 		if (!can_piggyback(pvc, cip, target_threads)) {
2468 			spin_unlock(&pvc->lock);
2469 			continue;
2470 		}
2471 		kvmppc_core_end_stolen(pvc);
2472 		pvc->vcore_state = VCORE_PIGGYBACK;
2473 		if (cip->total_threads >= target_threads)
2474 			break;
2475 	}
2476 	spin_unlock(&lp->lock);
2477 }
2478 
2479 static bool recheck_signals(struct core_info *cip)
2480 {
2481 	int sub, i;
2482 	struct kvm_vcpu *vcpu;
2483 
2484 	for (sub = 0; sub < cip->n_subcores; ++sub)
2485 		for_each_runnable_thread(i, vcpu, cip->vc[sub])
2486 			if (signal_pending(vcpu->arch.run_task))
2487 				return true;
2488 	return false;
2489 }
2490 
2491 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2492 {
2493 	int still_running = 0, i;
2494 	u64 now;
2495 	long ret;
2496 	struct kvm_vcpu *vcpu;
2497 
2498 	spin_lock(&vc->lock);
2499 	now = get_tb();
2500 	for_each_runnable_thread(i, vcpu, vc) {
2501 		/* cancel pending dec exception if dec is positive */
2502 		if (now < vcpu->arch.dec_expires &&
2503 		    kvmppc_core_pending_dec(vcpu))
2504 			kvmppc_core_dequeue_dec(vcpu);
2505 
2506 		trace_kvm_guest_exit(vcpu);
2507 
2508 		ret = RESUME_GUEST;
2509 		if (vcpu->arch.trap)
2510 			ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2511 						    vcpu->arch.run_task);
2512 
2513 		vcpu->arch.ret = ret;
2514 		vcpu->arch.trap = 0;
2515 
2516 		if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2517 			if (vcpu->arch.pending_exceptions)
2518 				kvmppc_core_prepare_to_enter(vcpu);
2519 			if (vcpu->arch.ceded)
2520 				kvmppc_set_timer(vcpu);
2521 			else
2522 				++still_running;
2523 		} else {
2524 			kvmppc_remove_runnable(vc, vcpu);
2525 			wake_up(&vcpu->arch.cpu_run);
2526 		}
2527 	}
2528 	if (!is_master) {
2529 		if (still_running > 0) {
2530 			kvmppc_vcore_preempt(vc);
2531 		} else if (vc->runner) {
2532 			vc->vcore_state = VCORE_PREEMPT;
2533 			kvmppc_core_start_stolen(vc);
2534 		} else {
2535 			vc->vcore_state = VCORE_INACTIVE;
2536 		}
2537 		if (vc->n_runnable > 0 && vc->runner == NULL) {
2538 			/* make sure there's a candidate runner awake */
2539 			i = -1;
2540 			vcpu = next_runnable_thread(vc, &i);
2541 			wake_up(&vcpu->arch.cpu_run);
2542 		}
2543 	}
2544 	spin_unlock(&vc->lock);
2545 }
2546 
2547 /*
2548  * Clear core from the list of active host cores as we are about to
2549  * enter the guest. Only do this if it is the primary thread of the
2550  * core (not if a subcore) that is entering the guest.
2551  */
2552 static inline int kvmppc_clear_host_core(unsigned int cpu)
2553 {
2554 	int core;
2555 
2556 	if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2557 		return 0;
2558 	/*
2559 	 * Memory barrier can be omitted here as we will do a smp_wmb()
2560 	 * later in kvmppc_start_thread and we need ensure that state is
2561 	 * visible to other CPUs only after we enter guest.
2562 	 */
2563 	core = cpu >> threads_shift;
2564 	kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
2565 	return 0;
2566 }
2567 
2568 /*
2569  * Advertise this core as an active host core since we exited the guest
2570  * Only need to do this if it is the primary thread of the core that is
2571  * exiting.
2572  */
2573 static inline int kvmppc_set_host_core(unsigned int cpu)
2574 {
2575 	int core;
2576 
2577 	if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2578 		return 0;
2579 
2580 	/*
2581 	 * Memory barrier can be omitted here because we do a spin_unlock
2582 	 * immediately after this which provides the memory barrier.
2583 	 */
2584 	core = cpu >> threads_shift;
2585 	kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
2586 	return 0;
2587 }
2588 
2589 static void set_irq_happened(int trap)
2590 {
2591 	switch (trap) {
2592 	case BOOK3S_INTERRUPT_EXTERNAL:
2593 		local_paca->irq_happened |= PACA_IRQ_EE;
2594 		break;
2595 	case BOOK3S_INTERRUPT_H_DOORBELL:
2596 		local_paca->irq_happened |= PACA_IRQ_DBELL;
2597 		break;
2598 	case BOOK3S_INTERRUPT_HMI:
2599 		local_paca->irq_happened |= PACA_IRQ_HMI;
2600 		break;
2601 	}
2602 }
2603 
2604 /*
2605  * Run a set of guest threads on a physical core.
2606  * Called with vc->lock held.
2607  */
2608 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
2609 {
2610 	struct kvm_vcpu *vcpu;
2611 	int i;
2612 	int srcu_idx;
2613 	struct core_info core_info;
2614 	struct kvmppc_vcore *pvc;
2615 	struct kvm_split_mode split_info, *sip;
2616 	int split, subcore_size, active;
2617 	int sub;
2618 	bool thr0_done;
2619 	unsigned long cmd_bit, stat_bit;
2620 	int pcpu, thr;
2621 	int target_threads;
2622 	int controlled_threads;
2623 	int trap;
2624 
2625 	/*
2626 	 * Remove from the list any threads that have a signal pending
2627 	 * or need a VPA update done
2628 	 */
2629 	prepare_threads(vc);
2630 
2631 	/* if the runner is no longer runnable, let the caller pick a new one */
2632 	if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
2633 		return;
2634 
2635 	/*
2636 	 * Initialize *vc.
2637 	 */
2638 	init_vcore_to_run(vc);
2639 	vc->preempt_tb = TB_NIL;
2640 
2641 	/*
2642 	 * Number of threads that we will be controlling: the same as
2643 	 * the number of threads per subcore, except on POWER9,
2644 	 * where it's 1 because the threads are (mostly) independent.
2645 	 */
2646 	controlled_threads = threads_per_vcore();
2647 
2648 	/*
2649 	 * Make sure we are running on primary threads, and that secondary
2650 	 * threads are offline.  Also check if the number of threads in this
2651 	 * guest are greater than the current system threads per guest.
2652 	 */
2653 	if ((controlled_threads > 1) &&
2654 	    ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
2655 		for_each_runnable_thread(i, vcpu, vc) {
2656 			vcpu->arch.ret = -EBUSY;
2657 			kvmppc_remove_runnable(vc, vcpu);
2658 			wake_up(&vcpu->arch.cpu_run);
2659 		}
2660 		goto out;
2661 	}
2662 
2663 	/*
2664 	 * See if we could run any other vcores on the physical core
2665 	 * along with this one.
2666 	 */
2667 	init_core_info(&core_info, vc);
2668 	pcpu = smp_processor_id();
2669 	target_threads = controlled_threads;
2670 	if (target_smt_mode && target_smt_mode < target_threads)
2671 		target_threads = target_smt_mode;
2672 	if (vc->num_threads < target_threads)
2673 		collect_piggybacks(&core_info, target_threads);
2674 
2675 	/*
2676 	 * On radix, arrange for TLB flushing if necessary.
2677 	 * This has to be done before disabling interrupts since
2678 	 * it uses smp_call_function().
2679 	 */
2680 	pcpu = smp_processor_id();
2681 	if (kvm_is_radix(vc->kvm)) {
2682 		for (sub = 0; sub < core_info.n_subcores; ++sub)
2683 			for_each_runnable_thread(i, vcpu, core_info.vc[sub])
2684 				kvmppc_prepare_radix_vcpu(vcpu, pcpu);
2685 	}
2686 
2687 	/*
2688 	 * Hard-disable interrupts, and check resched flag and signals.
2689 	 * If we need to reschedule or deliver a signal, clean up
2690 	 * and return without going into the guest(s).
2691 	 */
2692 	local_irq_disable();
2693 	hard_irq_disable();
2694 	if (lazy_irq_pending() || need_resched() ||
2695 	    recheck_signals(&core_info)) {
2696 		local_irq_enable();
2697 		vc->vcore_state = VCORE_INACTIVE;
2698 		/* Unlock all except the primary vcore */
2699 		for (sub = 1; sub < core_info.n_subcores; ++sub) {
2700 			pvc = core_info.vc[sub];
2701 			/* Put back on to the preempted vcores list */
2702 			kvmppc_vcore_preempt(pvc);
2703 			spin_unlock(&pvc->lock);
2704 		}
2705 		for (i = 0; i < controlled_threads; ++i)
2706 			kvmppc_release_hwthread(pcpu + i);
2707 		return;
2708 	}
2709 
2710 	kvmppc_clear_host_core(pcpu);
2711 
2712 	/* Decide on micro-threading (split-core) mode */
2713 	subcore_size = threads_per_subcore;
2714 	cmd_bit = stat_bit = 0;
2715 	split = core_info.n_subcores;
2716 	sip = NULL;
2717 	if (split > 1) {
2718 		/* threads_per_subcore must be MAX_SMT_THREADS (8) here */
2719 		if (split == 2 && (dynamic_mt_modes & 2)) {
2720 			cmd_bit = HID0_POWER8_1TO2LPAR;
2721 			stat_bit = HID0_POWER8_2LPARMODE;
2722 		} else {
2723 			split = 4;
2724 			cmd_bit = HID0_POWER8_1TO4LPAR;
2725 			stat_bit = HID0_POWER8_4LPARMODE;
2726 		}
2727 		subcore_size = MAX_SMT_THREADS / split;
2728 		sip = &split_info;
2729 		memset(&split_info, 0, sizeof(split_info));
2730 		split_info.rpr = mfspr(SPRN_RPR);
2731 		split_info.pmmar = mfspr(SPRN_PMMAR);
2732 		split_info.ldbar = mfspr(SPRN_LDBAR);
2733 		split_info.subcore_size = subcore_size;
2734 		for (sub = 0; sub < core_info.n_subcores; ++sub)
2735 			split_info.vc[sub] = core_info.vc[sub];
2736 		/* order writes to split_info before kvm_split_mode pointer */
2737 		smp_wmb();
2738 	}
2739 	for (thr = 0; thr < controlled_threads; ++thr)
2740 		paca[pcpu + thr].kvm_hstate.kvm_split_mode = sip;
2741 
2742 	/* Initiate micro-threading (split-core) if required */
2743 	if (cmd_bit) {
2744 		unsigned long hid0 = mfspr(SPRN_HID0);
2745 
2746 		hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
2747 		mb();
2748 		mtspr(SPRN_HID0, hid0);
2749 		isync();
2750 		for (;;) {
2751 			hid0 = mfspr(SPRN_HID0);
2752 			if (hid0 & stat_bit)
2753 				break;
2754 			cpu_relax();
2755 		}
2756 	}
2757 
2758 	/* Start all the threads */
2759 	active = 0;
2760 	for (sub = 0; sub < core_info.n_subcores; ++sub) {
2761 		thr = subcore_thread_map[sub];
2762 		thr0_done = false;
2763 		active |= 1 << thr;
2764 		pvc = core_info.vc[sub];
2765 		pvc->pcpu = pcpu + thr;
2766 		for_each_runnable_thread(i, vcpu, pvc) {
2767 			kvmppc_start_thread(vcpu, pvc);
2768 			kvmppc_create_dtl_entry(vcpu, pvc);
2769 			trace_kvm_guest_enter(vcpu);
2770 			if (!vcpu->arch.ptid)
2771 				thr0_done = true;
2772 			active |= 1 << (thr + vcpu->arch.ptid);
2773 		}
2774 		/*
2775 		 * We need to start the first thread of each subcore
2776 		 * even if it doesn't have a vcpu.
2777 		 */
2778 		if (!thr0_done)
2779 			kvmppc_start_thread(NULL, pvc);
2780 		thr += pvc->num_threads;
2781 	}
2782 
2783 	/*
2784 	 * Ensure that split_info.do_nap is set after setting
2785 	 * the vcore pointer in the PACA of the secondaries.
2786 	 */
2787 	smp_mb();
2788 	if (cmd_bit)
2789 		split_info.do_nap = 1;	/* ask secondaries to nap when done */
2790 
2791 	/*
2792 	 * When doing micro-threading, poke the inactive threads as well.
2793 	 * This gets them to the nap instruction after kvm_do_nap,
2794 	 * which reduces the time taken to unsplit later.
2795 	 */
2796 	if (split > 1)
2797 		for (thr = 1; thr < threads_per_subcore; ++thr)
2798 			if (!(active & (1 << thr)))
2799 				kvmppc_ipi_thread(pcpu + thr);
2800 
2801 	vc->vcore_state = VCORE_RUNNING;
2802 	preempt_disable();
2803 
2804 	trace_kvmppc_run_core(vc, 0);
2805 
2806 	for (sub = 0; sub < core_info.n_subcores; ++sub)
2807 		spin_unlock(&core_info.vc[sub]->lock);
2808 
2809 	/*
2810 	 * Interrupts will be enabled once we get into the guest,
2811 	 * so tell lockdep that we're about to enable interrupts.
2812 	 */
2813 	trace_hardirqs_on();
2814 
2815 	guest_enter();
2816 
2817 	srcu_idx = srcu_read_lock(&vc->kvm->srcu);
2818 
2819 	trap = __kvmppc_vcore_entry();
2820 
2821 	srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
2822 
2823 	guest_exit();
2824 
2825 	trace_hardirqs_off();
2826 	set_irq_happened(trap);
2827 
2828 	spin_lock(&vc->lock);
2829 	/* prevent other vcpu threads from doing kvmppc_start_thread() now */
2830 	vc->vcore_state = VCORE_EXITING;
2831 
2832 	/* wait for secondary threads to finish writing their state to memory */
2833 	kvmppc_wait_for_nap();
2834 
2835 	/* Return to whole-core mode if we split the core earlier */
2836 	if (split > 1) {
2837 		unsigned long hid0 = mfspr(SPRN_HID0);
2838 		unsigned long loops = 0;
2839 
2840 		hid0 &= ~HID0_POWER8_DYNLPARDIS;
2841 		stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
2842 		mb();
2843 		mtspr(SPRN_HID0, hid0);
2844 		isync();
2845 		for (;;) {
2846 			hid0 = mfspr(SPRN_HID0);
2847 			if (!(hid0 & stat_bit))
2848 				break;
2849 			cpu_relax();
2850 			++loops;
2851 		}
2852 		split_info.do_nap = 0;
2853 	}
2854 
2855 	kvmppc_set_host_core(pcpu);
2856 
2857 	local_irq_enable();
2858 
2859 	/* Let secondaries go back to the offline loop */
2860 	for (i = 0; i < controlled_threads; ++i) {
2861 		kvmppc_release_hwthread(pcpu + i);
2862 		if (sip && sip->napped[i])
2863 			kvmppc_ipi_thread(pcpu + i);
2864 		cpumask_clear_cpu(pcpu + i, &vc->kvm->arch.cpu_in_guest);
2865 	}
2866 
2867 	spin_unlock(&vc->lock);
2868 
2869 	/* make sure updates to secondary vcpu structs are visible now */
2870 	smp_mb();
2871 
2872 	for (sub = 0; sub < core_info.n_subcores; ++sub) {
2873 		pvc = core_info.vc[sub];
2874 		post_guest_process(pvc, pvc == vc);
2875 	}
2876 
2877 	spin_lock(&vc->lock);
2878 	preempt_enable();
2879 
2880  out:
2881 	vc->vcore_state = VCORE_INACTIVE;
2882 	trace_kvmppc_run_core(vc, 1);
2883 }
2884 
2885 /*
2886  * Wait for some other vcpu thread to execute us, and
2887  * wake us up when we need to handle something in the host.
2888  */
2889 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
2890 				 struct kvm_vcpu *vcpu, int wait_state)
2891 {
2892 	DEFINE_WAIT(wait);
2893 
2894 	prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
2895 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
2896 		spin_unlock(&vc->lock);
2897 		schedule();
2898 		spin_lock(&vc->lock);
2899 	}
2900 	finish_wait(&vcpu->arch.cpu_run, &wait);
2901 }
2902 
2903 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
2904 {
2905 	/* 10us base */
2906 	if (vc->halt_poll_ns == 0 && halt_poll_ns_grow)
2907 		vc->halt_poll_ns = 10000;
2908 	else
2909 		vc->halt_poll_ns *= halt_poll_ns_grow;
2910 }
2911 
2912 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
2913 {
2914 	if (halt_poll_ns_shrink == 0)
2915 		vc->halt_poll_ns = 0;
2916 	else
2917 		vc->halt_poll_ns /= halt_poll_ns_shrink;
2918 }
2919 
2920 #ifdef CONFIG_KVM_XICS
2921 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
2922 {
2923 	if (!xive_enabled())
2924 		return false;
2925 	return vcpu->arch.xive_saved_state.pipr <
2926 		vcpu->arch.xive_saved_state.cppr;
2927 }
2928 #else
2929 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
2930 {
2931 	return false;
2932 }
2933 #endif /* CONFIG_KVM_XICS */
2934 
2935 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
2936 {
2937 	if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
2938 	    kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
2939 		return true;
2940 
2941 	return false;
2942 }
2943 
2944 /*
2945  * Check to see if any of the runnable vcpus on the vcore have pending
2946  * exceptions or are no longer ceded
2947  */
2948 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
2949 {
2950 	struct kvm_vcpu *vcpu;
2951 	int i;
2952 
2953 	for_each_runnable_thread(i, vcpu, vc) {
2954 		if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
2955 			return 1;
2956 	}
2957 
2958 	return 0;
2959 }
2960 
2961 /*
2962  * All the vcpus in this vcore are idle, so wait for a decrementer
2963  * or external interrupt to one of the vcpus.  vc->lock is held.
2964  */
2965 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
2966 {
2967 	ktime_t cur, start_poll, start_wait;
2968 	int do_sleep = 1;
2969 	u64 block_ns;
2970 	DECLARE_SWAITQUEUE(wait);
2971 
2972 	/* Poll for pending exceptions and ceded state */
2973 	cur = start_poll = ktime_get();
2974 	if (vc->halt_poll_ns) {
2975 		ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
2976 		++vc->runner->stat.halt_attempted_poll;
2977 
2978 		vc->vcore_state = VCORE_POLLING;
2979 		spin_unlock(&vc->lock);
2980 
2981 		do {
2982 			if (kvmppc_vcore_check_block(vc)) {
2983 				do_sleep = 0;
2984 				break;
2985 			}
2986 			cur = ktime_get();
2987 		} while (single_task_running() && ktime_before(cur, stop));
2988 
2989 		spin_lock(&vc->lock);
2990 		vc->vcore_state = VCORE_INACTIVE;
2991 
2992 		if (!do_sleep) {
2993 			++vc->runner->stat.halt_successful_poll;
2994 			goto out;
2995 		}
2996 	}
2997 
2998 	prepare_to_swait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
2999 
3000 	if (kvmppc_vcore_check_block(vc)) {
3001 		finish_swait(&vc->wq, &wait);
3002 		do_sleep = 0;
3003 		/* If we polled, count this as a successful poll */
3004 		if (vc->halt_poll_ns)
3005 			++vc->runner->stat.halt_successful_poll;
3006 		goto out;
3007 	}
3008 
3009 	start_wait = ktime_get();
3010 
3011 	vc->vcore_state = VCORE_SLEEPING;
3012 	trace_kvmppc_vcore_blocked(vc, 0);
3013 	spin_unlock(&vc->lock);
3014 	schedule();
3015 	finish_swait(&vc->wq, &wait);
3016 	spin_lock(&vc->lock);
3017 	vc->vcore_state = VCORE_INACTIVE;
3018 	trace_kvmppc_vcore_blocked(vc, 1);
3019 	++vc->runner->stat.halt_successful_wait;
3020 
3021 	cur = ktime_get();
3022 
3023 out:
3024 	block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
3025 
3026 	/* Attribute wait time */
3027 	if (do_sleep) {
3028 		vc->runner->stat.halt_wait_ns +=
3029 			ktime_to_ns(cur) - ktime_to_ns(start_wait);
3030 		/* Attribute failed poll time */
3031 		if (vc->halt_poll_ns)
3032 			vc->runner->stat.halt_poll_fail_ns +=
3033 				ktime_to_ns(start_wait) -
3034 				ktime_to_ns(start_poll);
3035 	} else {
3036 		/* Attribute successful poll time */
3037 		if (vc->halt_poll_ns)
3038 			vc->runner->stat.halt_poll_success_ns +=
3039 				ktime_to_ns(cur) -
3040 				ktime_to_ns(start_poll);
3041 	}
3042 
3043 	/* Adjust poll time */
3044 	if (halt_poll_ns) {
3045 		if (block_ns <= vc->halt_poll_ns)
3046 			;
3047 		/* We slept and blocked for longer than the max halt time */
3048 		else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
3049 			shrink_halt_poll_ns(vc);
3050 		/* We slept and our poll time is too small */
3051 		else if (vc->halt_poll_ns < halt_poll_ns &&
3052 				block_ns < halt_poll_ns)
3053 			grow_halt_poll_ns(vc);
3054 		if (vc->halt_poll_ns > halt_poll_ns)
3055 			vc->halt_poll_ns = halt_poll_ns;
3056 	} else
3057 		vc->halt_poll_ns = 0;
3058 
3059 	trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
3060 }
3061 
3062 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
3063 {
3064 	int n_ceded, i;
3065 	struct kvmppc_vcore *vc;
3066 	struct kvm_vcpu *v;
3067 
3068 	trace_kvmppc_run_vcpu_enter(vcpu);
3069 
3070 	kvm_run->exit_reason = 0;
3071 	vcpu->arch.ret = RESUME_GUEST;
3072 	vcpu->arch.trap = 0;
3073 	kvmppc_update_vpas(vcpu);
3074 
3075 	/*
3076 	 * Synchronize with other threads in this virtual core
3077 	 */
3078 	vc = vcpu->arch.vcore;
3079 	spin_lock(&vc->lock);
3080 	vcpu->arch.ceded = 0;
3081 	vcpu->arch.run_task = current;
3082 	vcpu->arch.kvm_run = kvm_run;
3083 	vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
3084 	vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
3085 	vcpu->arch.busy_preempt = TB_NIL;
3086 	WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
3087 	++vc->n_runnable;
3088 
3089 	/*
3090 	 * This happens the first time this is called for a vcpu.
3091 	 * If the vcore is already running, we may be able to start
3092 	 * this thread straight away and have it join in.
3093 	 */
3094 	if (!signal_pending(current)) {
3095 		if (vc->vcore_state == VCORE_PIGGYBACK) {
3096 			if (spin_trylock(&vc->lock)) {
3097 				if (vc->vcore_state == VCORE_RUNNING &&
3098 				    !VCORE_IS_EXITING(vc)) {
3099 					kvmppc_create_dtl_entry(vcpu, vc);
3100 					kvmppc_start_thread(vcpu, vc);
3101 					trace_kvm_guest_enter(vcpu);
3102 				}
3103 				spin_unlock(&vc->lock);
3104 			}
3105 		} else if (vc->vcore_state == VCORE_RUNNING &&
3106 			   !VCORE_IS_EXITING(vc)) {
3107 			kvmppc_create_dtl_entry(vcpu, vc);
3108 			kvmppc_start_thread(vcpu, vc);
3109 			trace_kvm_guest_enter(vcpu);
3110 		} else if (vc->vcore_state == VCORE_SLEEPING) {
3111 			swake_up(&vc->wq);
3112 		}
3113 
3114 	}
3115 
3116 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3117 	       !signal_pending(current)) {
3118 		if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3119 			kvmppc_vcore_end_preempt(vc);
3120 
3121 		if (vc->vcore_state != VCORE_INACTIVE) {
3122 			kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
3123 			continue;
3124 		}
3125 		for_each_runnable_thread(i, v, vc) {
3126 			kvmppc_core_prepare_to_enter(v);
3127 			if (signal_pending(v->arch.run_task)) {
3128 				kvmppc_remove_runnable(vc, v);
3129 				v->stat.signal_exits++;
3130 				v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
3131 				v->arch.ret = -EINTR;
3132 				wake_up(&v->arch.cpu_run);
3133 			}
3134 		}
3135 		if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
3136 			break;
3137 		n_ceded = 0;
3138 		for_each_runnable_thread(i, v, vc) {
3139 			if (!kvmppc_vcpu_woken(v))
3140 				n_ceded += v->arch.ceded;
3141 			else
3142 				v->arch.ceded = 0;
3143 		}
3144 		vc->runner = vcpu;
3145 		if (n_ceded == vc->n_runnable) {
3146 			kvmppc_vcore_blocked(vc);
3147 		} else if (need_resched()) {
3148 			kvmppc_vcore_preempt(vc);
3149 			/* Let something else run */
3150 			cond_resched_lock(&vc->lock);
3151 			if (vc->vcore_state == VCORE_PREEMPT)
3152 				kvmppc_vcore_end_preempt(vc);
3153 		} else {
3154 			kvmppc_run_core(vc);
3155 		}
3156 		vc->runner = NULL;
3157 	}
3158 
3159 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3160 	       (vc->vcore_state == VCORE_RUNNING ||
3161 		vc->vcore_state == VCORE_EXITING ||
3162 		vc->vcore_state == VCORE_PIGGYBACK))
3163 		kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
3164 
3165 	if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3166 		kvmppc_vcore_end_preempt(vc);
3167 
3168 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3169 		kvmppc_remove_runnable(vc, vcpu);
3170 		vcpu->stat.signal_exits++;
3171 		kvm_run->exit_reason = KVM_EXIT_INTR;
3172 		vcpu->arch.ret = -EINTR;
3173 	}
3174 
3175 	if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
3176 		/* Wake up some vcpu to run the core */
3177 		i = -1;
3178 		v = next_runnable_thread(vc, &i);
3179 		wake_up(&v->arch.cpu_run);
3180 	}
3181 
3182 	trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
3183 	spin_unlock(&vc->lock);
3184 	return vcpu->arch.ret;
3185 }
3186 
3187 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
3188 {
3189 	int r;
3190 	int srcu_idx;
3191 	unsigned long ebb_regs[3] = {};	/* shut up GCC */
3192 	unsigned long user_tar = 0;
3193 	unsigned int user_vrsave;
3194 
3195 	if (!vcpu->arch.sane) {
3196 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3197 		return -EINVAL;
3198 	}
3199 
3200 	/*
3201 	 * Don't allow entry with a suspended transaction, because
3202 	 * the guest entry/exit code will lose it.
3203 	 * If the guest has TM enabled, save away their TM-related SPRs
3204 	 * (they will get restored by the TM unavailable interrupt).
3205 	 */
3206 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
3207 	if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
3208 	    (current->thread.regs->msr & MSR_TM)) {
3209 		if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
3210 			run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3211 			run->fail_entry.hardware_entry_failure_reason = 0;
3212 			return -EINVAL;
3213 		}
3214 		/* Enable TM so we can read the TM SPRs */
3215 		mtmsr(mfmsr() | MSR_TM);
3216 		current->thread.tm_tfhar = mfspr(SPRN_TFHAR);
3217 		current->thread.tm_tfiar = mfspr(SPRN_TFIAR);
3218 		current->thread.tm_texasr = mfspr(SPRN_TEXASR);
3219 		current->thread.regs->msr &= ~MSR_TM;
3220 	}
3221 #endif
3222 
3223 	kvmppc_core_prepare_to_enter(vcpu);
3224 
3225 	/* No need to go into the guest when all we'll do is come back out */
3226 	if (signal_pending(current)) {
3227 		run->exit_reason = KVM_EXIT_INTR;
3228 		return -EINTR;
3229 	}
3230 
3231 	atomic_inc(&vcpu->kvm->arch.vcpus_running);
3232 	/* Order vcpus_running vs. hpte_setup_done, see kvmppc_alloc_reset_hpt */
3233 	smp_mb();
3234 
3235 	/* On the first time here, set up HTAB and VRMA */
3236 	if (!kvm_is_radix(vcpu->kvm) && !vcpu->kvm->arch.hpte_setup_done) {
3237 		r = kvmppc_hv_setup_htab_rma(vcpu);
3238 		if (r)
3239 			goto out;
3240 	}
3241 
3242 	flush_all_to_thread(current);
3243 
3244 	/* Save userspace EBB and other register values */
3245 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
3246 		ebb_regs[0] = mfspr(SPRN_EBBHR);
3247 		ebb_regs[1] = mfspr(SPRN_EBBRR);
3248 		ebb_regs[2] = mfspr(SPRN_BESCR);
3249 		user_tar = mfspr(SPRN_TAR);
3250 	}
3251 	user_vrsave = mfspr(SPRN_VRSAVE);
3252 
3253 	vcpu->arch.wqp = &vcpu->arch.vcore->wq;
3254 	vcpu->arch.pgdir = current->mm->pgd;
3255 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
3256 
3257 	do {
3258 		r = kvmppc_run_vcpu(run, vcpu);
3259 
3260 		if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
3261 		    !(vcpu->arch.shregs.msr & MSR_PR)) {
3262 			trace_kvm_hcall_enter(vcpu);
3263 			r = kvmppc_pseries_do_hcall(vcpu);
3264 			trace_kvm_hcall_exit(vcpu, r);
3265 			kvmppc_core_prepare_to_enter(vcpu);
3266 		} else if (r == RESUME_PAGE_FAULT) {
3267 			srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
3268 			r = kvmppc_book3s_hv_page_fault(run, vcpu,
3269 				vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
3270 			srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
3271 		} else if (r == RESUME_PASSTHROUGH) {
3272 			if (WARN_ON(xive_enabled()))
3273 				r = H_SUCCESS;
3274 			else
3275 				r = kvmppc_xics_rm_complete(vcpu, 0);
3276 		}
3277 	} while (is_kvmppc_resume_guest(r));
3278 
3279 	/* Restore userspace EBB and other register values */
3280 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
3281 		mtspr(SPRN_EBBHR, ebb_regs[0]);
3282 		mtspr(SPRN_EBBRR, ebb_regs[1]);
3283 		mtspr(SPRN_BESCR, ebb_regs[2]);
3284 		mtspr(SPRN_TAR, user_tar);
3285 		mtspr(SPRN_FSCR, current->thread.fscr);
3286 	}
3287 	mtspr(SPRN_VRSAVE, user_vrsave);
3288 
3289  out:
3290 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
3291 	atomic_dec(&vcpu->kvm->arch.vcpus_running);
3292 	return r;
3293 }
3294 
3295 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
3296 				     int linux_psize)
3297 {
3298 	struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
3299 
3300 	if (!def->shift)
3301 		return;
3302 	(*sps)->page_shift = def->shift;
3303 	(*sps)->slb_enc = def->sllp;
3304 	(*sps)->enc[0].page_shift = def->shift;
3305 	(*sps)->enc[0].pte_enc = def->penc[linux_psize];
3306 	/*
3307 	 * Add 16MB MPSS support if host supports it
3308 	 */
3309 	if (linux_psize != MMU_PAGE_16M && def->penc[MMU_PAGE_16M] != -1) {
3310 		(*sps)->enc[1].page_shift = 24;
3311 		(*sps)->enc[1].pte_enc = def->penc[MMU_PAGE_16M];
3312 	}
3313 	(*sps)++;
3314 }
3315 
3316 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
3317 					 struct kvm_ppc_smmu_info *info)
3318 {
3319 	struct kvm_ppc_one_seg_page_size *sps;
3320 
3321 	/*
3322 	 * Since we don't yet support HPT guests on a radix host,
3323 	 * return an error if the host uses radix.
3324 	 */
3325 	if (radix_enabled())
3326 		return -EINVAL;
3327 
3328 	info->flags = KVM_PPC_PAGE_SIZES_REAL;
3329 	if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
3330 		info->flags |= KVM_PPC_1T_SEGMENTS;
3331 	info->slb_size = mmu_slb_size;
3332 
3333 	/* We only support these sizes for now, and no muti-size segments */
3334 	sps = &info->sps[0];
3335 	kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
3336 	kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
3337 	kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
3338 
3339 	return 0;
3340 }
3341 
3342 /*
3343  * Get (and clear) the dirty memory log for a memory slot.
3344  */
3345 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
3346 					 struct kvm_dirty_log *log)
3347 {
3348 	struct kvm_memslots *slots;
3349 	struct kvm_memory_slot *memslot;
3350 	int i, r;
3351 	unsigned long n;
3352 	unsigned long *buf;
3353 	struct kvm_vcpu *vcpu;
3354 
3355 	mutex_lock(&kvm->slots_lock);
3356 
3357 	r = -EINVAL;
3358 	if (log->slot >= KVM_USER_MEM_SLOTS)
3359 		goto out;
3360 
3361 	slots = kvm_memslots(kvm);
3362 	memslot = id_to_memslot(slots, log->slot);
3363 	r = -ENOENT;
3364 	if (!memslot->dirty_bitmap)
3365 		goto out;
3366 
3367 	/*
3368 	 * Use second half of bitmap area because radix accumulates
3369 	 * bits in the first half.
3370 	 */
3371 	n = kvm_dirty_bitmap_bytes(memslot);
3372 	buf = memslot->dirty_bitmap + n / sizeof(long);
3373 	memset(buf, 0, n);
3374 
3375 	if (kvm_is_radix(kvm))
3376 		r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
3377 	else
3378 		r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
3379 	if (r)
3380 		goto out;
3381 
3382 	/* Harvest dirty bits from VPA and DTL updates */
3383 	/* Note: we never modify the SLB shadow buffer areas */
3384 	kvm_for_each_vcpu(i, vcpu, kvm) {
3385 		spin_lock(&vcpu->arch.vpa_update_lock);
3386 		kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
3387 		kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
3388 		spin_unlock(&vcpu->arch.vpa_update_lock);
3389 	}
3390 
3391 	r = -EFAULT;
3392 	if (copy_to_user(log->dirty_bitmap, buf, n))
3393 		goto out;
3394 
3395 	r = 0;
3396 out:
3397 	mutex_unlock(&kvm->slots_lock);
3398 	return r;
3399 }
3400 
3401 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
3402 					struct kvm_memory_slot *dont)
3403 {
3404 	if (!dont || free->arch.rmap != dont->arch.rmap) {
3405 		vfree(free->arch.rmap);
3406 		free->arch.rmap = NULL;
3407 	}
3408 }
3409 
3410 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
3411 					 unsigned long npages)
3412 {
3413 	/*
3414 	 * For now, if radix_enabled() then we only support radix guests,
3415 	 * and in that case we don't need the rmap array.
3416 	 */
3417 	if (radix_enabled()) {
3418 		slot->arch.rmap = NULL;
3419 		return 0;
3420 	}
3421 
3422 	slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
3423 	if (!slot->arch.rmap)
3424 		return -ENOMEM;
3425 
3426 	return 0;
3427 }
3428 
3429 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
3430 					struct kvm_memory_slot *memslot,
3431 					const struct kvm_userspace_memory_region *mem)
3432 {
3433 	return 0;
3434 }
3435 
3436 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
3437 				const struct kvm_userspace_memory_region *mem,
3438 				const struct kvm_memory_slot *old,
3439 				const struct kvm_memory_slot *new)
3440 {
3441 	unsigned long npages = mem->memory_size >> PAGE_SHIFT;
3442 	struct kvm_memslots *slots;
3443 	struct kvm_memory_slot *memslot;
3444 
3445 	/*
3446 	 * If we are making a new memslot, it might make
3447 	 * some address that was previously cached as emulated
3448 	 * MMIO be no longer emulated MMIO, so invalidate
3449 	 * all the caches of emulated MMIO translations.
3450 	 */
3451 	if (npages)
3452 		atomic64_inc(&kvm->arch.mmio_update);
3453 
3454 	if (npages && old->npages && !kvm_is_radix(kvm)) {
3455 		/*
3456 		 * If modifying a memslot, reset all the rmap dirty bits.
3457 		 * If this is a new memslot, we don't need to do anything
3458 		 * since the rmap array starts out as all zeroes,
3459 		 * i.e. no pages are dirty.
3460 		 */
3461 		slots = kvm_memslots(kvm);
3462 		memslot = id_to_memslot(slots, mem->slot);
3463 		kvmppc_hv_get_dirty_log_hpt(kvm, memslot, NULL);
3464 	}
3465 }
3466 
3467 /*
3468  * Update LPCR values in kvm->arch and in vcores.
3469  * Caller must hold kvm->lock.
3470  */
3471 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
3472 {
3473 	long int i;
3474 	u32 cores_done = 0;
3475 
3476 	if ((kvm->arch.lpcr & mask) == lpcr)
3477 		return;
3478 
3479 	kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
3480 
3481 	for (i = 0; i < KVM_MAX_VCORES; ++i) {
3482 		struct kvmppc_vcore *vc = kvm->arch.vcores[i];
3483 		if (!vc)
3484 			continue;
3485 		spin_lock(&vc->lock);
3486 		vc->lpcr = (vc->lpcr & ~mask) | lpcr;
3487 		spin_unlock(&vc->lock);
3488 		if (++cores_done >= kvm->arch.online_vcores)
3489 			break;
3490 	}
3491 }
3492 
3493 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
3494 {
3495 	return;
3496 }
3497 
3498 static void kvmppc_setup_partition_table(struct kvm *kvm)
3499 {
3500 	unsigned long dw0, dw1;
3501 
3502 	if (!kvm_is_radix(kvm)) {
3503 		/* PS field - page size for VRMA */
3504 		dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
3505 			((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
3506 		/* HTABSIZE and HTABORG fields */
3507 		dw0 |= kvm->arch.sdr1;
3508 
3509 		/* Second dword as set by userspace */
3510 		dw1 = kvm->arch.process_table;
3511 	} else {
3512 		dw0 = PATB_HR | radix__get_tree_size() |
3513 			__pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
3514 		dw1 = PATB_GR | kvm->arch.process_table;
3515 	}
3516 
3517 	mmu_partition_table_set_entry(kvm->arch.lpid, dw0, dw1);
3518 }
3519 
3520 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
3521 {
3522 	int err = 0;
3523 	struct kvm *kvm = vcpu->kvm;
3524 	unsigned long hva;
3525 	struct kvm_memory_slot *memslot;
3526 	struct vm_area_struct *vma;
3527 	unsigned long lpcr = 0, senc;
3528 	unsigned long psize, porder;
3529 	int srcu_idx;
3530 
3531 	mutex_lock(&kvm->lock);
3532 	if (kvm->arch.hpte_setup_done)
3533 		goto out;	/* another vcpu beat us to it */
3534 
3535 	/* Allocate hashed page table (if not done already) and reset it */
3536 	if (!kvm->arch.hpt.virt) {
3537 		int order = KVM_DEFAULT_HPT_ORDER;
3538 		struct kvm_hpt_info info;
3539 
3540 		err = kvmppc_allocate_hpt(&info, order);
3541 		/* If we get here, it means userspace didn't specify a
3542 		 * size explicitly.  So, try successively smaller
3543 		 * sizes if the default failed. */
3544 		while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
3545 			err  = kvmppc_allocate_hpt(&info, order);
3546 
3547 		if (err < 0) {
3548 			pr_err("KVM: Couldn't alloc HPT\n");
3549 			goto out;
3550 		}
3551 
3552 		kvmppc_set_hpt(kvm, &info);
3553 	}
3554 
3555 	/* Look up the memslot for guest physical address 0 */
3556 	srcu_idx = srcu_read_lock(&kvm->srcu);
3557 	memslot = gfn_to_memslot(kvm, 0);
3558 
3559 	/* We must have some memory at 0 by now */
3560 	err = -EINVAL;
3561 	if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
3562 		goto out_srcu;
3563 
3564 	/* Look up the VMA for the start of this memory slot */
3565 	hva = memslot->userspace_addr;
3566 	down_read(&current->mm->mmap_sem);
3567 	vma = find_vma(current->mm, hva);
3568 	if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
3569 		goto up_out;
3570 
3571 	psize = vma_kernel_pagesize(vma);
3572 	porder = __ilog2(psize);
3573 
3574 	up_read(&current->mm->mmap_sem);
3575 
3576 	/* We can handle 4k, 64k or 16M pages in the VRMA */
3577 	err = -EINVAL;
3578 	if (!(psize == 0x1000 || psize == 0x10000 ||
3579 	      psize == 0x1000000))
3580 		goto out_srcu;
3581 
3582 	senc = slb_pgsize_encoding(psize);
3583 	kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
3584 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
3585 	/* Create HPTEs in the hash page table for the VRMA */
3586 	kvmppc_map_vrma(vcpu, memslot, porder);
3587 
3588 	/* Update VRMASD field in the LPCR */
3589 	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
3590 		/* the -4 is to account for senc values starting at 0x10 */
3591 		lpcr = senc << (LPCR_VRMASD_SH - 4);
3592 		kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
3593 	} else {
3594 		kvmppc_setup_partition_table(kvm);
3595 	}
3596 
3597 	/* Order updates to kvm->arch.lpcr etc. vs. hpte_setup_done */
3598 	smp_wmb();
3599 	kvm->arch.hpte_setup_done = 1;
3600 	err = 0;
3601  out_srcu:
3602 	srcu_read_unlock(&kvm->srcu, srcu_idx);
3603  out:
3604 	mutex_unlock(&kvm->lock);
3605 	return err;
3606 
3607  up_out:
3608 	up_read(&current->mm->mmap_sem);
3609 	goto out_srcu;
3610 }
3611 
3612 #ifdef CONFIG_KVM_XICS
3613 /*
3614  * Allocate a per-core structure for managing state about which cores are
3615  * running in the host versus the guest and for exchanging data between
3616  * real mode KVM and CPU running in the host.
3617  * This is only done for the first VM.
3618  * The allocated structure stays even if all VMs have stopped.
3619  * It is only freed when the kvm-hv module is unloaded.
3620  * It's OK for this routine to fail, we just don't support host
3621  * core operations like redirecting H_IPI wakeups.
3622  */
3623 void kvmppc_alloc_host_rm_ops(void)
3624 {
3625 	struct kvmppc_host_rm_ops *ops;
3626 	unsigned long l_ops;
3627 	int cpu, core;
3628 	int size;
3629 
3630 	/* Not the first time here ? */
3631 	if (kvmppc_host_rm_ops_hv != NULL)
3632 		return;
3633 
3634 	ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
3635 	if (!ops)
3636 		return;
3637 
3638 	size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
3639 	ops->rm_core = kzalloc(size, GFP_KERNEL);
3640 
3641 	if (!ops->rm_core) {
3642 		kfree(ops);
3643 		return;
3644 	}
3645 
3646 	cpus_read_lock();
3647 
3648 	for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
3649 		if (!cpu_online(cpu))
3650 			continue;
3651 
3652 		core = cpu >> threads_shift;
3653 		ops->rm_core[core].rm_state.in_host = 1;
3654 	}
3655 
3656 	ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
3657 
3658 	/*
3659 	 * Make the contents of the kvmppc_host_rm_ops structure visible
3660 	 * to other CPUs before we assign it to the global variable.
3661 	 * Do an atomic assignment (no locks used here), but if someone
3662 	 * beats us to it, just free our copy and return.
3663 	 */
3664 	smp_wmb();
3665 	l_ops = (unsigned long) ops;
3666 
3667 	if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
3668 		cpus_read_unlock();
3669 		kfree(ops->rm_core);
3670 		kfree(ops);
3671 		return;
3672 	}
3673 
3674 	cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
3675 					     "ppc/kvm_book3s:prepare",
3676 					     kvmppc_set_host_core,
3677 					     kvmppc_clear_host_core);
3678 	cpus_read_unlock();
3679 }
3680 
3681 void kvmppc_free_host_rm_ops(void)
3682 {
3683 	if (kvmppc_host_rm_ops_hv) {
3684 		cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
3685 		kfree(kvmppc_host_rm_ops_hv->rm_core);
3686 		kfree(kvmppc_host_rm_ops_hv);
3687 		kvmppc_host_rm_ops_hv = NULL;
3688 	}
3689 }
3690 #endif
3691 
3692 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
3693 {
3694 	unsigned long lpcr, lpid;
3695 	char buf[32];
3696 	int ret;
3697 
3698 	/* Allocate the guest's logical partition ID */
3699 
3700 	lpid = kvmppc_alloc_lpid();
3701 	if ((long)lpid < 0)
3702 		return -ENOMEM;
3703 	kvm->arch.lpid = lpid;
3704 
3705 	kvmppc_alloc_host_rm_ops();
3706 
3707 	/*
3708 	 * Since we don't flush the TLB when tearing down a VM,
3709 	 * and this lpid might have previously been used,
3710 	 * make sure we flush on each core before running the new VM.
3711 	 * On POWER9, the tlbie in mmu_partition_table_set_entry()
3712 	 * does this flush for us.
3713 	 */
3714 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
3715 		cpumask_setall(&kvm->arch.need_tlb_flush);
3716 
3717 	/* Start out with the default set of hcalls enabled */
3718 	memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
3719 	       sizeof(kvm->arch.enabled_hcalls));
3720 
3721 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
3722 		kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
3723 
3724 	/* Init LPCR for virtual RMA mode */
3725 	kvm->arch.host_lpid = mfspr(SPRN_LPID);
3726 	kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
3727 	lpcr &= LPCR_PECE | LPCR_LPES;
3728 	lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
3729 		LPCR_VPM0 | LPCR_VPM1;
3730 	kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
3731 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
3732 	/* On POWER8 turn on online bit to enable PURR/SPURR */
3733 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
3734 		lpcr |= LPCR_ONL;
3735 	/*
3736 	 * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
3737 	 * Set HVICE bit to enable hypervisor virtualization interrupts.
3738 	 * Set HEIC to prevent OS interrupts to go to hypervisor (should
3739 	 * be unnecessary but better safe than sorry in case we re-enable
3740 	 * EE in HV mode with this LPCR still set)
3741 	 */
3742 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
3743 		lpcr &= ~LPCR_VPM0;
3744 		lpcr |= LPCR_HVICE | LPCR_HEIC;
3745 
3746 		/*
3747 		 * If xive is enabled, we route 0x500 interrupts directly
3748 		 * to the guest.
3749 		 */
3750 		if (xive_enabled())
3751 			lpcr |= LPCR_LPES;
3752 	}
3753 
3754 	/*
3755 	 * For now, if the host uses radix, the guest must be radix.
3756 	 */
3757 	if (radix_enabled()) {
3758 		kvm->arch.radix = 1;
3759 		lpcr &= ~LPCR_VPM1;
3760 		lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
3761 		ret = kvmppc_init_vm_radix(kvm);
3762 		if (ret) {
3763 			kvmppc_free_lpid(kvm->arch.lpid);
3764 			return ret;
3765 		}
3766 		kvmppc_setup_partition_table(kvm);
3767 	}
3768 
3769 	kvm->arch.lpcr = lpcr;
3770 
3771 	/* Initialization for future HPT resizes */
3772 	kvm->arch.resize_hpt = NULL;
3773 
3774 	/*
3775 	 * Work out how many sets the TLB has, for the use of
3776 	 * the TLB invalidation loop in book3s_hv_rmhandlers.S.
3777 	 */
3778 	if (kvm_is_radix(kvm))
3779 		kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX;	/* 128 */
3780 	else if (cpu_has_feature(CPU_FTR_ARCH_300))
3781 		kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH;	/* 256 */
3782 	else if (cpu_has_feature(CPU_FTR_ARCH_207S))
3783 		kvm->arch.tlb_sets = POWER8_TLB_SETS;		/* 512 */
3784 	else
3785 		kvm->arch.tlb_sets = POWER7_TLB_SETS;		/* 128 */
3786 
3787 	/*
3788 	 * Track that we now have a HV mode VM active. This blocks secondary
3789 	 * CPU threads from coming online.
3790 	 * On POWER9, we only need to do this for HPT guests on a radix
3791 	 * host, which is not yet supported.
3792 	 */
3793 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
3794 		kvm_hv_vm_activated();
3795 
3796 	/*
3797 	 * Initialize smt_mode depending on processor.
3798 	 * POWER8 and earlier have to use "strict" threading, where
3799 	 * all vCPUs in a vcore have to run on the same (sub)core,
3800 	 * whereas on POWER9 the threads can each run a different
3801 	 * guest.
3802 	 */
3803 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
3804 		kvm->arch.smt_mode = threads_per_subcore;
3805 	else
3806 		kvm->arch.smt_mode = 1;
3807 	kvm->arch.emul_smt_mode = 1;
3808 
3809 	/*
3810 	 * Create a debugfs directory for the VM
3811 	 */
3812 	snprintf(buf, sizeof(buf), "vm%d", current->pid);
3813 	kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
3814 	if (!IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
3815 		kvmppc_mmu_debugfs_init(kvm);
3816 
3817 	return 0;
3818 }
3819 
3820 static void kvmppc_free_vcores(struct kvm *kvm)
3821 {
3822 	long int i;
3823 
3824 	for (i = 0; i < KVM_MAX_VCORES; ++i)
3825 		kfree(kvm->arch.vcores[i]);
3826 	kvm->arch.online_vcores = 0;
3827 }
3828 
3829 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
3830 {
3831 	debugfs_remove_recursive(kvm->arch.debugfs_dir);
3832 
3833 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
3834 		kvm_hv_vm_deactivated();
3835 
3836 	kvmppc_free_vcores(kvm);
3837 
3838 	kvmppc_free_lpid(kvm->arch.lpid);
3839 
3840 	if (kvm_is_radix(kvm))
3841 		kvmppc_free_radix(kvm);
3842 	else
3843 		kvmppc_free_hpt(&kvm->arch.hpt);
3844 
3845 	kvmppc_free_pimap(kvm);
3846 }
3847 
3848 /* We don't need to emulate any privileged instructions or dcbz */
3849 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
3850 				     unsigned int inst, int *advance)
3851 {
3852 	return EMULATE_FAIL;
3853 }
3854 
3855 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
3856 					ulong spr_val)
3857 {
3858 	return EMULATE_FAIL;
3859 }
3860 
3861 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
3862 					ulong *spr_val)
3863 {
3864 	return EMULATE_FAIL;
3865 }
3866 
3867 static int kvmppc_core_check_processor_compat_hv(void)
3868 {
3869 	if (!cpu_has_feature(CPU_FTR_HVMODE) ||
3870 	    !cpu_has_feature(CPU_FTR_ARCH_206))
3871 		return -EIO;
3872 
3873 	return 0;
3874 }
3875 
3876 #ifdef CONFIG_KVM_XICS
3877 
3878 void kvmppc_free_pimap(struct kvm *kvm)
3879 {
3880 	kfree(kvm->arch.pimap);
3881 }
3882 
3883 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
3884 {
3885 	return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
3886 }
3887 
3888 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
3889 {
3890 	struct irq_desc *desc;
3891 	struct kvmppc_irq_map *irq_map;
3892 	struct kvmppc_passthru_irqmap *pimap;
3893 	struct irq_chip *chip;
3894 	int i, rc = 0;
3895 
3896 	if (!kvm_irq_bypass)
3897 		return 1;
3898 
3899 	desc = irq_to_desc(host_irq);
3900 	if (!desc)
3901 		return -EIO;
3902 
3903 	mutex_lock(&kvm->lock);
3904 
3905 	pimap = kvm->arch.pimap;
3906 	if (pimap == NULL) {
3907 		/* First call, allocate structure to hold IRQ map */
3908 		pimap = kvmppc_alloc_pimap();
3909 		if (pimap == NULL) {
3910 			mutex_unlock(&kvm->lock);
3911 			return -ENOMEM;
3912 		}
3913 		kvm->arch.pimap = pimap;
3914 	}
3915 
3916 	/*
3917 	 * For now, we only support interrupts for which the EOI operation
3918 	 * is an OPAL call followed by a write to XIRR, since that's
3919 	 * what our real-mode EOI code does, or a XIVE interrupt
3920 	 */
3921 	chip = irq_data_get_irq_chip(&desc->irq_data);
3922 	if (!chip || !(is_pnv_opal_msi(chip) || is_xive_irq(chip))) {
3923 		pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
3924 			host_irq, guest_gsi);
3925 		mutex_unlock(&kvm->lock);
3926 		return -ENOENT;
3927 	}
3928 
3929 	/*
3930 	 * See if we already have an entry for this guest IRQ number.
3931 	 * If it's mapped to a hardware IRQ number, that's an error,
3932 	 * otherwise re-use this entry.
3933 	 */
3934 	for (i = 0; i < pimap->n_mapped; i++) {
3935 		if (guest_gsi == pimap->mapped[i].v_hwirq) {
3936 			if (pimap->mapped[i].r_hwirq) {
3937 				mutex_unlock(&kvm->lock);
3938 				return -EINVAL;
3939 			}
3940 			break;
3941 		}
3942 	}
3943 
3944 	if (i == KVMPPC_PIRQ_MAPPED) {
3945 		mutex_unlock(&kvm->lock);
3946 		return -EAGAIN;		/* table is full */
3947 	}
3948 
3949 	irq_map = &pimap->mapped[i];
3950 
3951 	irq_map->v_hwirq = guest_gsi;
3952 	irq_map->desc = desc;
3953 
3954 	/*
3955 	 * Order the above two stores before the next to serialize with
3956 	 * the KVM real mode handler.
3957 	 */
3958 	smp_wmb();
3959 	irq_map->r_hwirq = desc->irq_data.hwirq;
3960 
3961 	if (i == pimap->n_mapped)
3962 		pimap->n_mapped++;
3963 
3964 	if (xive_enabled())
3965 		rc = kvmppc_xive_set_mapped(kvm, guest_gsi, desc);
3966 	else
3967 		kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq);
3968 	if (rc)
3969 		irq_map->r_hwirq = 0;
3970 
3971 	mutex_unlock(&kvm->lock);
3972 
3973 	return 0;
3974 }
3975 
3976 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
3977 {
3978 	struct irq_desc *desc;
3979 	struct kvmppc_passthru_irqmap *pimap;
3980 	int i, rc = 0;
3981 
3982 	if (!kvm_irq_bypass)
3983 		return 0;
3984 
3985 	desc = irq_to_desc(host_irq);
3986 	if (!desc)
3987 		return -EIO;
3988 
3989 	mutex_lock(&kvm->lock);
3990 	if (!kvm->arch.pimap)
3991 		goto unlock;
3992 
3993 	pimap = kvm->arch.pimap;
3994 
3995 	for (i = 0; i < pimap->n_mapped; i++) {
3996 		if (guest_gsi == pimap->mapped[i].v_hwirq)
3997 			break;
3998 	}
3999 
4000 	if (i == pimap->n_mapped) {
4001 		mutex_unlock(&kvm->lock);
4002 		return -ENODEV;
4003 	}
4004 
4005 	if (xive_enabled())
4006 		rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, pimap->mapped[i].desc);
4007 	else
4008 		kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
4009 
4010 	/* invalidate the entry (what do do on error from the above ?) */
4011 	pimap->mapped[i].r_hwirq = 0;
4012 
4013 	/*
4014 	 * We don't free this structure even when the count goes to
4015 	 * zero. The structure is freed when we destroy the VM.
4016 	 */
4017  unlock:
4018 	mutex_unlock(&kvm->lock);
4019 	return rc;
4020 }
4021 
4022 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
4023 					     struct irq_bypass_producer *prod)
4024 {
4025 	int ret = 0;
4026 	struct kvm_kernel_irqfd *irqfd =
4027 		container_of(cons, struct kvm_kernel_irqfd, consumer);
4028 
4029 	irqfd->producer = prod;
4030 
4031 	ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
4032 	if (ret)
4033 		pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
4034 			prod->irq, irqfd->gsi, ret);
4035 
4036 	return ret;
4037 }
4038 
4039 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
4040 					      struct irq_bypass_producer *prod)
4041 {
4042 	int ret;
4043 	struct kvm_kernel_irqfd *irqfd =
4044 		container_of(cons, struct kvm_kernel_irqfd, consumer);
4045 
4046 	irqfd->producer = NULL;
4047 
4048 	/*
4049 	 * When producer of consumer is unregistered, we change back to
4050 	 * default external interrupt handling mode - KVM real mode
4051 	 * will switch back to host.
4052 	 */
4053 	ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
4054 	if (ret)
4055 		pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
4056 			prod->irq, irqfd->gsi, ret);
4057 }
4058 #endif
4059 
4060 static long kvm_arch_vm_ioctl_hv(struct file *filp,
4061 				 unsigned int ioctl, unsigned long arg)
4062 {
4063 	struct kvm *kvm __maybe_unused = filp->private_data;
4064 	void __user *argp = (void __user *)arg;
4065 	long r;
4066 
4067 	switch (ioctl) {
4068 
4069 	case KVM_PPC_ALLOCATE_HTAB: {
4070 		u32 htab_order;
4071 
4072 		r = -EFAULT;
4073 		if (get_user(htab_order, (u32 __user *)argp))
4074 			break;
4075 		r = kvmppc_alloc_reset_hpt(kvm, htab_order);
4076 		if (r)
4077 			break;
4078 		r = 0;
4079 		break;
4080 	}
4081 
4082 	case KVM_PPC_GET_HTAB_FD: {
4083 		struct kvm_get_htab_fd ghf;
4084 
4085 		r = -EFAULT;
4086 		if (copy_from_user(&ghf, argp, sizeof(ghf)))
4087 			break;
4088 		r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
4089 		break;
4090 	}
4091 
4092 	case KVM_PPC_RESIZE_HPT_PREPARE: {
4093 		struct kvm_ppc_resize_hpt rhpt;
4094 
4095 		r = -EFAULT;
4096 		if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
4097 			break;
4098 
4099 		r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
4100 		break;
4101 	}
4102 
4103 	case KVM_PPC_RESIZE_HPT_COMMIT: {
4104 		struct kvm_ppc_resize_hpt rhpt;
4105 
4106 		r = -EFAULT;
4107 		if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
4108 			break;
4109 
4110 		r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
4111 		break;
4112 	}
4113 
4114 	default:
4115 		r = -ENOTTY;
4116 	}
4117 
4118 	return r;
4119 }
4120 
4121 /*
4122  * List of hcall numbers to enable by default.
4123  * For compatibility with old userspace, we enable by default
4124  * all hcalls that were implemented before the hcall-enabling
4125  * facility was added.  Note this list should not include H_RTAS.
4126  */
4127 static unsigned int default_hcall_list[] = {
4128 	H_REMOVE,
4129 	H_ENTER,
4130 	H_READ,
4131 	H_PROTECT,
4132 	H_BULK_REMOVE,
4133 	H_GET_TCE,
4134 	H_PUT_TCE,
4135 	H_SET_DABR,
4136 	H_SET_XDABR,
4137 	H_CEDE,
4138 	H_PROD,
4139 	H_CONFER,
4140 	H_REGISTER_VPA,
4141 #ifdef CONFIG_KVM_XICS
4142 	H_EOI,
4143 	H_CPPR,
4144 	H_IPI,
4145 	H_IPOLL,
4146 	H_XIRR,
4147 	H_XIRR_X,
4148 #endif
4149 	0
4150 };
4151 
4152 static void init_default_hcalls(void)
4153 {
4154 	int i;
4155 	unsigned int hcall;
4156 
4157 	for (i = 0; default_hcall_list[i]; ++i) {
4158 		hcall = default_hcall_list[i];
4159 		WARN_ON(!kvmppc_hcall_impl_hv(hcall));
4160 		__set_bit(hcall / 4, default_enabled_hcalls);
4161 	}
4162 }
4163 
4164 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
4165 {
4166 	unsigned long lpcr;
4167 	int radix;
4168 
4169 	/* If not on a POWER9, reject it */
4170 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
4171 		return -ENODEV;
4172 
4173 	/* If any unknown flags set, reject it */
4174 	if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
4175 		return -EINVAL;
4176 
4177 	/* We can't change a guest to/from radix yet */
4178 	radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
4179 	if (radix != kvm_is_radix(kvm))
4180 		return -EINVAL;
4181 
4182 	/* GR (guest radix) bit in process_table field must match */
4183 	if (!!(cfg->process_table & PATB_GR) != radix)
4184 		return -EINVAL;
4185 
4186 	/* Process table size field must be reasonable, i.e. <= 24 */
4187 	if ((cfg->process_table & PRTS_MASK) > 24)
4188 		return -EINVAL;
4189 
4190 	kvm->arch.process_table = cfg->process_table;
4191 	kvmppc_setup_partition_table(kvm);
4192 
4193 	lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
4194 	kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
4195 
4196 	return 0;
4197 }
4198 
4199 static struct kvmppc_ops kvm_ops_hv = {
4200 	.get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
4201 	.set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
4202 	.get_one_reg = kvmppc_get_one_reg_hv,
4203 	.set_one_reg = kvmppc_set_one_reg_hv,
4204 	.vcpu_load   = kvmppc_core_vcpu_load_hv,
4205 	.vcpu_put    = kvmppc_core_vcpu_put_hv,
4206 	.set_msr     = kvmppc_set_msr_hv,
4207 	.vcpu_run    = kvmppc_vcpu_run_hv,
4208 	.vcpu_create = kvmppc_core_vcpu_create_hv,
4209 	.vcpu_free   = kvmppc_core_vcpu_free_hv,
4210 	.check_requests = kvmppc_core_check_requests_hv,
4211 	.get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
4212 	.flush_memslot  = kvmppc_core_flush_memslot_hv,
4213 	.prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
4214 	.commit_memory_region  = kvmppc_core_commit_memory_region_hv,
4215 	.unmap_hva = kvm_unmap_hva_hv,
4216 	.unmap_hva_range = kvm_unmap_hva_range_hv,
4217 	.age_hva  = kvm_age_hva_hv,
4218 	.test_age_hva = kvm_test_age_hva_hv,
4219 	.set_spte_hva = kvm_set_spte_hva_hv,
4220 	.mmu_destroy  = kvmppc_mmu_destroy_hv,
4221 	.free_memslot = kvmppc_core_free_memslot_hv,
4222 	.create_memslot = kvmppc_core_create_memslot_hv,
4223 	.init_vm =  kvmppc_core_init_vm_hv,
4224 	.destroy_vm = kvmppc_core_destroy_vm_hv,
4225 	.get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
4226 	.emulate_op = kvmppc_core_emulate_op_hv,
4227 	.emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
4228 	.emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
4229 	.fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
4230 	.arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
4231 	.hcall_implemented = kvmppc_hcall_impl_hv,
4232 #ifdef CONFIG_KVM_XICS
4233 	.irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
4234 	.irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
4235 #endif
4236 	.configure_mmu = kvmhv_configure_mmu,
4237 	.get_rmmu_info = kvmhv_get_rmmu_info,
4238 	.set_smt_mode = kvmhv_set_smt_mode,
4239 };
4240 
4241 static int kvm_init_subcore_bitmap(void)
4242 {
4243 	int i, j;
4244 	int nr_cores = cpu_nr_cores();
4245 	struct sibling_subcore_state *sibling_subcore_state;
4246 
4247 	for (i = 0; i < nr_cores; i++) {
4248 		int first_cpu = i * threads_per_core;
4249 		int node = cpu_to_node(first_cpu);
4250 
4251 		/* Ignore if it is already allocated. */
4252 		if (paca[first_cpu].sibling_subcore_state)
4253 			continue;
4254 
4255 		sibling_subcore_state =
4256 			kmalloc_node(sizeof(struct sibling_subcore_state),
4257 							GFP_KERNEL, node);
4258 		if (!sibling_subcore_state)
4259 			return -ENOMEM;
4260 
4261 		memset(sibling_subcore_state, 0,
4262 				sizeof(struct sibling_subcore_state));
4263 
4264 		for (j = 0; j < threads_per_core; j++) {
4265 			int cpu = first_cpu + j;
4266 
4267 			paca[cpu].sibling_subcore_state = sibling_subcore_state;
4268 		}
4269 	}
4270 	return 0;
4271 }
4272 
4273 static int kvmppc_radix_possible(void)
4274 {
4275 	return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
4276 }
4277 
4278 static int kvmppc_book3s_init_hv(void)
4279 {
4280 	int r;
4281 	/*
4282 	 * FIXME!! Do we need to check on all cpus ?
4283 	 */
4284 	r = kvmppc_core_check_processor_compat_hv();
4285 	if (r < 0)
4286 		return -ENODEV;
4287 
4288 	r = kvm_init_subcore_bitmap();
4289 	if (r)
4290 		return r;
4291 
4292 	/*
4293 	 * We need a way of accessing the XICS interrupt controller,
4294 	 * either directly, via paca[cpu].kvm_hstate.xics_phys, or
4295 	 * indirectly, via OPAL.
4296 	 */
4297 #ifdef CONFIG_SMP
4298 	if (!xive_enabled() && !local_paca->kvm_hstate.xics_phys) {
4299 		struct device_node *np;
4300 
4301 		np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
4302 		if (!np) {
4303 			pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
4304 			return -ENODEV;
4305 		}
4306 	}
4307 #endif
4308 
4309 	kvm_ops_hv.owner = THIS_MODULE;
4310 	kvmppc_hv_ops = &kvm_ops_hv;
4311 
4312 	init_default_hcalls();
4313 
4314 	init_vcore_lists();
4315 
4316 	r = kvmppc_mmu_hv_init();
4317 	if (r)
4318 		return r;
4319 
4320 	if (kvmppc_radix_possible())
4321 		r = kvmppc_radix_init();
4322 	return r;
4323 }
4324 
4325 static void kvmppc_book3s_exit_hv(void)
4326 {
4327 	kvmppc_free_host_rm_ops();
4328 	if (kvmppc_radix_possible())
4329 		kvmppc_radix_exit();
4330 	kvmppc_hv_ops = NULL;
4331 }
4332 
4333 module_init(kvmppc_book3s_init_hv);
4334 module_exit(kvmppc_book3s_exit_hv);
4335 MODULE_LICENSE("GPL");
4336 MODULE_ALIAS_MISCDEV(KVM_MINOR);
4337 MODULE_ALIAS("devname:kvm");
4338 
4339